Deeply Scaled CMOS for RF Power Applications by J*rg Scholvin B.S., Massachusetts Institute of Technology (2001) M.Eng., Massachusetts Institute of Technology (2001) Submitted to the Department of Electrical Engineering and Computer Science in Partial Fulfillment of the Requirements for the Degree of MASSACHUISETS INSTITUTE OF TECHNOLOGY Doctor of Philosophy NOV 0,2 2006 at the LIBRARIES MASSACHUSETTS INSTITUTE OF TECHNOLOGY June 2006 © 2006 Massachusetts Institute of Technology. All rights reserved. Author 1% I 0 s Department of Electrical Engineering and Computer Science May 23 2006 Certified by .. .. Jesuis A. del Alamo essor of Electrical Engineering Thesis Supervisor Certified by Arthur C. Smith Chairman, Department Committee on Graduate Thesis ARCHIVES Deeply Scaled CMOS for RF Power Applications by J6rg Scholvin Submitted to the Department of Electrical Engineering and Computer Science in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Electrical Engineering Abstract The microelectronics industry is striving to reduce the cost, complexity, and form factor of wireless systems through single-chip integration of analog, RF and digital functions. Driven by the requirements of the digital system components, the 90 nm and 65 nm technology nodes are currently emerging as platforms for highly integrated systems. Achieving such integration while minimizing the cost of adding specialized RF modules places high demands on the base CMOS technology. In this regard, the integration of the power amplifier (PA) function becomes an increasing challenge as technology geometries and supply voltages scale down. Gate length (Lg) scaling yields improved frequency response, promising higher power-added efficiency (PAE), a key RF PA consideration. This benefit comes at the cost of a lower drain voltage, which demands a higher output current and thus wider devices in order to produce a given output power level (Po,,). In this work, we have investigated the potential of deeply scaled CMOS for RF power applications, from 0.25 um down to 65 nm. We demonstrate the frequency and power limitations that the different CMOS technologies face, and describe the physical mechanisms that give rise to these limitations. We find that layout considerations, such as splitting a single large device into many smaller parallel devices, become increasingly important as the technology scales down the roadmap, both for power and frequency. We also show that parasitic resistances associated with the back-end wiring are responsible for placing an upper limit on the RF power that can be obtained for a single bond pad. We demonstrate a power density of 31 mW/mm for the 65 nm node, with PAE in excess of 60% at 4 GHz and 1 V. Similar results are obtained in 90 nm, where a peak PAE of 66% was measured at 2.2 GHz and 1 V, with a power density of 24 mW/mm. We find that efficient integrated PA functionality for many applications can be achieved even in a deeply-scaled logic CMOS technology. For low power levels (below 50 mW), we find that the 65 nm CMOS devices offer excellent efficiency (>50%) over a broad frequency range (2-8 GHz). Their RF power performance approaches that of 90 nm devices both in peak PAE and output power density. This is possible without costly PA-specific add-ons, or the use of higher voltage input-output (I/O) device options. However, since I/O devices are often included as part of the process, they represent a real option for PA integration because they allow for higher power densities. The 0.25 /im I/O device that is available in the 90 nm process, when biased at Vdd = 2.5 V showed excellent results, with a peak PAE of 60% and an output power of 75 mW (125 mW/mm) at 8 GHz. Thesis Supervisor: Jesuis A. del Alamo Title: Professor Acknowledgements The past several years working on this thesis have been an amazing experience of learning and del working with extraordinary people. I would like to thank my thesis advisor, Prof. Jesus and Alamo, for giving me the privilege to work in his group and to learn from him the skills been an knowledge I have acquired in the past years. Being able to learn from a great teacher has invaluable resource. for I would also like to thank my thesis readers, Prof. Joel Dawson and Prof. Charles Sodini thesis. I providing valuable guidance that helped me to improve the work presented in this deeply appreciate the feedback and suggestions that I have received. This work would not have been possible without the generous support of IBM, through the PhD del fellowship program and faculty awards. I would like to thank my mentor at IBM and former tireless Alamo student, David Greenberg, for making this collaboration possible through his Jack efforts and help. I also want to thank David Harame, Scott Parker, Greg Freeman and IBM at Pekarik for their enthusiastic support of the project. I have spent three summers working Burlington VT, and Fiskill NY, on projects directly related to this thesis. Among the many great device people I have met at IBM, I would like to thank Susan Sweeney and Jae Rascoe for their measurement expertise, and Ramana Malladi for many interesting conversations about my work. I want to thank everyone in the del Alamo research group for their support, inputs and outputs: former students Jim Fiorenza and Samuel Mertens, Niamh Waldron, Joyce Wu, Anita Villanueva, Melinda Wong, Jungwoo Joh, and post-doc's Tetsuya Suemitsu and Dae-Hyun Kim, as well as visiting scientist Joerg Appenzeller. They helped to make the office- and lab-life enjoyable. I would like to thank my wife Sylvie for being there for me and understanding the sometimes erratic work-schedule of a graduate student. And sometimes, great wisdom can be found in the most unexpected places: 0 A. ... ......... Contents 1. Introduction 1.1. Literature Background: State of the Art 1.2. Application Space for RF Power CMOS 1.3. Thesis Goals 1.4. Outline 2. Experimental 2.1. Technology Description 2.2. Device Layout 2.3. DC and S-Parameter Characterization 2.4. Power Characterization 3. 90 nm Technology 3.1. Results 3.1.1. RF Power Performance of the 90 nm Logic Device 3.1.2. RF Power Performance of Other Device Options 3.2. Discussion 3.3. Conclusion 4. 65 nm Technology 4.1. Experiments 4.2. Discussion 4.3. Technology Scaling: 65, 90 and 250 nm 5. First Order RF Power Model 5.1. Model Structure 5.2. Parameter Extraction 5.3. Model Results 6. Discussion 6.1. Frequency and Power Limits of 65 nm, 90 nm and 250 nm technologies 7. Conclusions Appendix A Power Measurement Theory 89 B Load-pull Setup 101 C Device Library 105 Bibliography 117 List of Figures 1.1 Number of power-amplifier research results published in IEEE journals and conferences in 1993 (top) and 2003 (bottom), by materialand device types. In 2003, 16% of the overall activity was taking place in CMOS technology. 1.2 Research activity as a function of time by material(left) and within silicon (right) by device type. 1.3 Power density as a function of bias voltage. Boundaries of different materials and device types are shown. CMOS is able to operate at voltages below 5 V, and demonstrates power densities comparable to the other materials. 19 Physical gate length of CMOS power devices and amplifiers, as a function of time, for IEEE published research. The trend towards scaling the gate exists even in RF power CMOS, as the data suggests. The CMOS technology nodes are marked by the arrows. Shorter gates generally result in higher gain (through transconductance)and therefore better efficiency and performance - though at the cost of reduced bias voltage and output power (see Figs. 1.3, 1.5). 19 Physicalgate length as a function of operationfrequency for CMOS RF power devices and amplifiers. Device operationat higherfrequencies requires a shortergate, as can be seen from the data. 17 18 1.4 1.5 20 1.6 Impact of gate length scaling on power added efficiency for class A to B devices, in a selected frequency band. Efficiency improves with shorter gates, due to higher transconductance and improved gain and frequency response. Data includes MESFET and HEMT type devices. 20 1.7 Overview of the different applications that require RF power. The sketch shows the broad range of frequencies and power levels that are required in different applications. 22 1.8 Technology landscape in the frequency-power space. Silicon based technologies occupy the low frequency regions, while compound semiconductors are used for higherfrequencies and power levels. 1.9 2.1 2.2 22 Illustration of how RF CMOS devices exhibit limitations of power andfrequency. The goal of this thesis is to identify these limits, and study how they can be pushed out. 26 Vertical drain layout (source = red, drain = blue, gate=green).The structure is symmetric, and only half is shown. The symmetry plane is indicated.The layout design is patentpending. 29 DC sweep for the standard90 nm device (1x48x16 pm), with Vgs from 0.4V to 1V in 0.1V steps. 30 2.3 2.4 S-parameters (0.5-50 GHz) for the standard90 nm device (1x48x16 pm), biased at Vds=1V, Vgs=0.58V. S11 and S22 are plotted on the left, and S21 and 50 times S12 are plotted on the right. 31 A typical power sweep, showing the power added efficiency (PAE) and the transducergain, as a function of the measured output power. This measurement shows the 65 nm standarddevice (768 Am width), at 4 GHz in class AB bias. 33 3.1 Power performance at 2.2 GHz of a thin-oxide 90 nm device (1x48x16 Am) and a device array (8x48x16 Am), both biased at Vdd = 1 V and Id = 26 mA/mm. Impedances were optimized for the best linearity-efficiency tradeoff. 37 3.2 Peak PAE and output power at peak PAE as a function offrequency for the standard90 nm device (1x48x16 Alm). All measurement points were taken at Vdd = I V, Id = 26 mA/mm, and impedances were reoptimized at eachfrequency. The output power is roughly independent of frequency, while the peak PAE is dropping as the frequency increases. 3.3 Gain, PAE and output power density at peak PAE point as a function of Vdd for thin oxide devices of three different gate lengths at 8 GHz. Impedances and input power drive were re-optimized at each Vddfor each device. 3.4 Gain, PAE and output power density at peak PAE as a function of Vddfor 250 nm devices with three different gate oxide thicknesses at 8 GHz. Impedances and input power drive were re-optimized at each Vddfor each device. 3.5 Gain, PAE and output power density at peak PAE point as a function of Vdd when scaling both L, and gate oxide thickness at 8 GHz. Impedances and input power drive were re-optimized at each Vddfor each device. 3.6 Comparison of the 90 nm thin and 250 nm thick oxide devices. Both devices can achieve similarpeak PAE levels, though at different operating voltages and power densities, as indicated. 3.7 Comparison off,, for the devices in Table I operating at Vnomina. Small width devices (right axis) with their pad parasiticsde-embedded offer a peak f, of 100 to 200 GHz. The f,. of the power devices (left axis) used in this paperare not de-embedded, and show a much lower f,, due to the class AB bias (Id = 26 mA/mm) and the parasiticsassociatedwith the pads, the large number offingers and their width (e.g. R, and Cgd). 3.8 Load-line comparisonfor the 90 nm thin and 250 nm thick oxide devices. MeasuredDC I-V characteristicsof the two devices are shown, with the approximate Vd,,sa indicated. The 90 nm device's Vds,sat is lower than that of the 250 nm device. Load-lines, based on the impedances used in the power measurements of Fig. 3.5, are also shown at Vdd = 1V and 2.5V. 38 40 41 42 42 43 45 4.1 4.2 Power characteristicsof two 65 nm devices with a total gate width of 768 lm at 4 GHz. The 12 cell device delivers a peak PAE of 65.7% at a power of 13.8 dBm (31.2 mW/mm) at a voltage of 1 V. 50 Power performancefor the 12x16x4 Alm, 65 nm device at 8 GHz vs. Vd. For each Vdd the device is tuned to peak PAE. The gain is the small signal gain. 51 4.3 Output characteristicsof the 12x16x4 pm, 65 nm device for V, = 0.2 to 1 V in 0.2 V steps. Forsmall values of Vgs and high values of Vds, impact ionization effect can be seen. Overlapping the device characteristicsis the load line correspondingto a biaspoint of I V. 4.4 Performanceof a 65 nm device (12x16x4 tm) as a function of bias current. Idq is the quiescent drain current (no Pin), and Id is the drain currentat peak PAE. The device is self biasing and the peak PAE, Pout and Gain do not change much with bias current. 4.5 Powerperformance as a function of device size for 65 nm devices (1, 2, 3, 4 and 6 parallelcells of 64x12 Alm each) and 90 nm devices (1, 2, 4 and 8 cells of 48x16 Atm). As the number of cells and thus device width increasesfor the 65 nm devices, PAE drops while Pout saturates. The behavior of 90 nm is very different as Pout continues to increase with device size. Given the independence of power on frequency over the range that is studied in this work (see Fig. 4.13), we have included a data pointfor a very wide 90 nm device (6.1 mm) that was measuredat 2.2 GHz. 52 52 53 4.6 4.7 4.8 4.9 4.10 4.11 4.12 4.13 4.14 5.1 (a, left) Simplified model schematic to create a multi-cell device from a number of single cell devices (b, right). The intrinsic device cells of the single cell and the multi-cell device are identical. Adding parallel single cell devices requires parasitic resistances to match the increased wire resistance of the multi-cell device's backend. The model parameterswere extracted using S-parametermeasurementsfrom a single cell and the multi-cell device. 54 Y11 comparison of a two-cell device (2x48x16 Am) with the model. When using two parallel 1x48x16 Am device measurements, the match with the two-cell device is mediocre. Adding parasiticresistances on the source, gate and drain,as shown in Fig. 4.6b, results in a much better match with the two-cell measurements. The other Y-parameters exhibit a similarpicture and are omitted to preserve clarity in the graph. 54 Small signal extracted parasitic scaling resistances (R, and Rs+Rd). To construct a multi-cell device from single-cell devices, these resistance must be added into the single cell unit (NFx WF = 48x12 ALm = 768 jm) as shown in Fig. 4.6b. 56 Model of the peak PAE (left) and Pou, performance (right) with scaling, matched to the 768 jm device, and using the resistances from Fig. 4.8. The graph also shows the ideal scaling behavior, as well as that of reducing the extracted resistance by 50%. 57 Correlation of peak PAE with f m,, (extracted at Vgs = 0.4 V and Vds= V) for the 5 different 65 nm device sizes (each cell is a 65 nm device of 64x12 im), as well as the optimized device layout (12 cells 16x4 jlm devices) 60 f,,, vs. unit finger width for a 1 cell device, keeping overall width the same (NF x WG,F = 768 jm). Large finger width hurts fm,a because of R,. Narrow finger width hurts because of distributedeffects involving the large number of device fingers that must be parallelized. 60 Breaking the 65 nm device up into multiple cells in parallel, both the number offingers and their unit width are scaled down at the same time. This method can achieve higherf,,, values (51.4 GHz vs. 43.7 GHz) at identical bias conditions. However, too many cells in parallel hurt performance due to the excessive wiring required,as seen by the drop in fm,x as the number of cells increases beyond 50. 61 Performance comparison of the 90 nm and 65 nm devices as a function of frequency. Both devices have a total width of :768 Am (65 nm: 64x12 Am and 90 nm: 48x16 jLm). The output power is independent of frequency, and essentially identical for both devices, while the peak PAE for this 65 nm device is slightly lower. 62 Peak PAE and output power density at 8 GHz as afunction of Vddfor three different CMOS technologies 64 DC characteristicsof the 90-nm device (1x48x16 Am) with indicated load-lines. The model mustfocus on the device regions that the load-line occupies. The DC data shows self heating effects for high Vds, Id values. 68 5.2 Full device model (top) and simplified equivalent model (bottom). The R-C networks on the drainand gate are simplified into a singlefrequency dependent resistor. The assumption is that the reactive components of the device are resonatedout by the source and load impedances. The other model elements include the currentgenerator(k, VT, exp), the breakdown voltage (BV) and the on-resistance(Ron). 69 5.3 Examplefit of the subthresholdcurve at Vds= IlV, for the 90 nm device. The extracted values for VT, k and exp are shown in Table 5.1. 71 5.4 Measuredoutput resistance (Real(Y22)) as a function of Vgs, for two frequencies at Vd= I V. The data shows that Ro is bias dependent. The model assumes a bias-independentRdo• Choosing a constant value around threshold (indicated)provides the best compromise. 72 5.5 Measuredinput resistance (Real(Y11)) as a function of Vg, for two frequenciesat Vds= 1 V. The data shows that Rgs is largely bias independent. Model parameters were chosen at V,= VT. 72 5.6 Model performance comparison with measuredpowerperformance, as a function offrequency. The model uses the parametersof Table 5.1. The model for each technology roughly matches the behavior of the data. The output power is frequency independentup to very high frequencies, while the peak PAE decreases. 74 5.7 6.1 6.2 6.3 6.4 7.1 Model performance compared to the 65 nm device scaling measurements. Depending on the parasitic resistancefor the overall device, the drop in peak PAE and the maximum output power can vary significantly. The model assumes a constant backend resistance that does not scale with the number of cells. 74 Constant PAE contours illustrating the output power that can be achieved at a given frequency. The higher the desired PAE, the lower the power and frequency. The power has an upper limit set by the backend parasitics,while the frequency has an upper limit related to the device gain andf,,• The values in the graph are based on the 90 nm technology, which we expect to be able of delivering 140 mW of power (Pat,) with a maximum power at Nc.sa,=12 , and a maximum operatingfrequency of 21 GHz (fa). 80 Impact of device layout on the device fn,,. The standardsingle cell device is shown in the solid bars. Using an optimized device layout by breaking the device up into multiple smaller cells, f,,a can be improved significantly. This becomes more critical as the technology is scaled down. 82 Estimated contour of 50% peak PAE, for three different technologies, illustrating the power vs. frequency behavior. The estimates are based on extrapolationof measureddata. The limited power of the 65 nm device in chapter 4 can be expected to improve with a different layout design (shown as dotted lines). 82 Estimated 50% peak PAE contours, from Fig. 6.2, redrawn taking the device layout optimization into account. Layout optimization can help to push out the frequency limit for the PA device. The 65 nm technology profitsfrom optimized layout the most, as seen in Fig. 6.3. 84 Expected 50% peak PAE contours, from Fig. 6.4. This figure assumes an optimized backend layout for the 65 nm devices, resulting in an improved width scaling. We expect the maximum power of the 65 nm devices to come close to that of the 90 nm devices. 88 List of Tables 2.1 3.1 3.2 4.1 5.1 Key device and process parametersfor thefour device types in this thesis 28 Gate length, oxide thickness and gate dimensionsfor the devices characterizedin this work. The device size refers to the number offingers x unitfinger width. Sizes have been chosen to give identical drive current for the 90 nm thin-, 130 nm medium-, and 250 nm thick-oxide devices. 36 Performance comparison of standard 90 nm device and a parallel combination of 8 standard devices. Impedances were optimized for the best linearity-efficiency tradeoff. 36 Overview of the devices used in this chapter,fabricated in either 65 or 90 nm technologies. The gate length refers to the equivalentpolysilicon halfpitch dimension. 49 Model parametersextractedfrom DC measurements,for the 65 nm and 90 nm technologies. 71 Chapter 1. Introduction This work will perform a detailed comparative study of the RF power potential of CMOS technology. The broader context of this research is the attractiveness of using logic CMOS as a platform for wireless system-on-chip applications. This enables true single-chip solutions for portable wireless systems with significant consequences in form factor and module design. Logic CMOS is the ideal platform for system-on-chip integration. Logic CMOS is the lowest cost technology, and the digital baseband in wireless systems is already being implemented in logic CMOS. CMOS modeling, simulation, design and verification tools are very mature and broadly available. Additionally, CMOS designs port relatively well from fab to fab and generation to generation. Particularly important for RF wireless systems, great progress has been shown recently in enabling RF functions with CMOS. Not surprisingly, there is a lot of interest in exploring the RF potential of logic CMOS. The microelectronics industry is striving to reduce the cost, complexity, and time to market of wireless systems through single-chip integration of mixed-signal RF/digital functions. Achieving such integration, while minimizing the increased cost of technology add-ons, places high demands on the base CMOS technology. In this regard, the integration of the power amplifier (PA) function remains a particular challenge as technology geometries scale down. Over the past decade, silicon technology has made a strong entry into the RF power amplifier field. Figure 1.1 shows the growth of silicon and CMOS in published research on device technologies for RF power amplifiers. In the commercial sector, CMOS has captured substantial market share for RF power amplifier applications. These include portable wireless systems, such as bluetooth and wireless LAN. The use of CMOS is because of cost and integration pressure. With cost a central issue in most high-volume applications, the digital portions of these applications will continue to be scaled down. As one example, the migration of CDMA chipsets onto 65 nm started in 2005. While this migration takes place only for the digital chips, it shows that any future integration requires the analog/RF chips to be implemented in deeply scaled technologies. Through a combination of detailed RF power characterization and analysis, in this thesis, we will study the RF power performance of deeply scaled CMOS, and determine its performance limits. We focus on conducting as opposed to switching type amplifiers. We will consider both the standard digital devices of each node, as well as available 1/O (thick gate oxide) devices. The answers to this study are necessarily complex because this is a multidimensional space: power level, frequency, linearity, power efficiency and reliability. 1.1. Literature Background: State of the Art To better illustrate the role of CMOS in RF power amplifiers, we have conducted a literature study of IEEE publications concerning power amplifiers operating in the RF above 900 MHz. Results from over 950 papers spanning the past 40 years were collected. Fig. 1.1 Number of power-amplifier research results published in IEEE journals and conferences in 1993 (top) and 2003 (bottom), by material and device types. In 2003, 16% of the overall activity was taking place in CMOS technology. [31-1006] The growing importance of silicon based power amplifiers can be seen in Figs. 1.1 and 1.2. Between 1993 and 2003, the portion of silicon based power amplifiers in publications has taken a substantial share of the overall PA research. CMOS based power amplifiers make up about 50% of the PA activity taking place in silicon today (Fig. 1.2b). Clearly offering the cheapest and most integratable technology, CMOS also is able to deliver performance at low voltages [1]. Fig. 1.3 shows the power density as a function of the bias voltage for a variety of semiconductor technologies. CMOS is competitive with respect to the other technologies such as GaAs HEMT and MESFETs, up to about 3 Volts. With the sharp increase in portable consumer electronics operating at frequencies below 5 GHz and voltages below 3V, CMOS has become an important candidate for integrated PAs, able to challenge competing technologies. Currently, research is mostly focused on circuit design techniques, such as linearization techniques and integrated transceivers [2-4]. Much less research has been done on CMOS technology itself [5-7]. This is the object of this work. The trend towards scaling of RF power CMOS is apparent from Figs 1.4-1.6. Over time, publications concerned with power CMOS have continued to follow the CMOS roadmap and used shorter gate lengths (Fig. 1.4). There are two important reasons why scaling makes sense. First, as the gate length is scaled down, operation at higher frequencies is possible (Fig. 1.5). In addition, scaling holds benefits even if the frequency of operation is not changed, as shown in Fig. 1.6. Due to the increase in transconductance and power gain that comes with shorter channels, scaled devices achieve higher power added efficiency and improved performance [1]. Inn - Research Activity in Silicon 40 > 75o. - 30- N Si BJT SiGe HBT LDMOS S CMOS SS S50- by Device Type - 20a- 250 10E 0- Year SI'#'ill Ill Fig. 1.2a (left), 1.2b (right) Research activity as a function of time by material (left) and within silicon (right) by device type. [31-1006] 4 r~r. IUU E 10 E Si , 0.1 - o 0.. 0.001 0 0 0.0001 ·· 0.1· 10 0.1 100 Vdd [V] Fig. 1.3 Power density as a function of bias voltage. Boundaries of different materialsand device types are shown. CMOS is able to operate at voltages below 5 V, and demonstrates power densities comparable to the other materials.[31-1006] IU.UU CMUS Scaling 1.00 0 o 0o o o S - 0.35 pm o *-- 0.25 pm -So- 0.10 0.01 1990 o - 1995 2000 0.18pm 0.13 pm 90 nm 2005 Year Fig. 1.4 Physical gate length of CMOS power devices and amplifiers, as a function of time, for IEEE published research. The trend towards scaling the gate exists even in RF power CMOS, as the data suggests. The CMOS technology nodes are marked by the arrows. Shorter gates generally result in higher gain (through transconductance) and therefore better efficiency and performance- though at the cost of reduced bias voltage and output power (see Fig. 1.3, 1.5). ~n nn IU.00 1.00 ¢• -. ~f-lg 0 "-. 06 4,, 0.10 o 0 4, o 0.01 4, · 1.0 10.0 Frequency [GHz] Fig. 1.5 Physical gate length as a function of operationfrequency for CMOS RF power devices and amplifiers. Device operation at higherfrequencies requires a shortergate, as can be seen from the data. 60 Impact of Scaling on Efficiency (27 GHz < freq < 50 GHz) - 8 .40 I6 - 4,1Q ( 20 - 4, S 0 i 0.0 0.2 0.4 0.6 0.8 1.0 LG [um] Fig. 1.6 Impact of gate length scaling on power added efficiency in a selected frequency band. Efficiency improves with shorter gates, due to higher transconductanceand improved gain and frequency response. Data includes MESFET and HEMT based amplifiers. 1.2. Application Space for Power Amplifiers Two of the most important requirements for a power amplifier are the output power level and the frequency of operation. We would like to illustrate the application demands on these two parameters. Fig. 1.7 shows this, by sketching the current RF power application landscape into the frequency-power space. The applications span 1100 GHz in frequency, and 10 mW to 100 W in output power level. Because of this broad range of frequencies and power levels, no single one technology dominates this space. In Fig. 1.8 we overlap technologies shown in the previous section (Fig. 1.1) onto this space. Silicon based technologies dominate the low frequency space, with CMOS occupying the low power, low frequency corner. LDMOS devices are suitable to higher power levels, and SiGe HBTs allow for higher frequencies. 100 10 1 0.1 AV.,A I '""''' ' ' ' """ ` """' 10 100 "~-"' 1000 Frequency [GHz] Fig. 1.7 Overview of the different applicationsthat require RF power. The sketch shows the broadrange offrequencies andpower levels that are requiredin different applications. 100 10 1 0.1 U.V I 10 100 1000 Frequency [GHz] Fig. 1.8 Technology landscape in the frequency-power space. Silicon based technologies occupy the low frequency regions, while compound semiconductors are used for higherfrequencies and power levels. 1.3. Thesis Goals This thesis is attempting to develop fundamental understanding about the RF power potential of logic CMOS. Through a combination of detailed RF power characterization and analysis, we wish to answer the following questions: * What is the RF power potential of logic CMOS? * How does the RF power performance of CMOS evolve as the technology is scaled down the road map? * What is the impact of device layout on RF power performance? The answers to these questions are necessarily complex because this is a multidimensional space: power level, frequency, power efficiency, layout considerations and device biasing choices. Fig. 1.9. illustrates this goal. We would like to identify the boundary of RF CMOS to accomplish the power amplifier function, both in terms of frequency and output power. In addition, the goal is to identify methods to push out these boundaries. We will accomplish the goals mentioned in the introduction by studying a variety of aspects which combined will give an excellent insight into RF power capability of deeply scaled CMOS: A Technology Our evaluation will focus on the 90 nm and 65 nm nodes. We will study different device options commonly available at these nodes: thin/medium/thick gate oxides (equivalent to 65 nm, 90 nm, 0.13 tm and 0.25 tpm devices), variation of gate length, and device layout variations. The analysis will include measurement of power and efficiency. This will then enable us to quantify the tradeoffs between the standard CMOS devices and more complex but potentially better performing alternative devices. B Technology Scaling By including various oxide thicknesses on a 90 nm wafer, we are essentially able to study devices of the 90 nm, 130 nm and 250 nm nodes. In addition, devices from a 65 nm wafer allow us to also study the 65 nm node. These two wafers give us a wide range of different technologies along the CMOS roadmap. The impact of scaling on power added efficiency and output power for conduction type amplifiers will be studied. C Layout Design We will develop new device layouts to improve RF power performance. Important considerations for developing RF power layouts are RF and DC parasitic concerns. The impact of the layouts on performance and frequency response will be investigated. D Power Capability We will study the ability of the technology to produce high output power by studying how device size relates to performance. The output power can saturate with device size for several reasons, such as gate resistance, self heating, inefficient power combining, and impedance mismatches. We will study a wide range of device widths that should allow us to identify the limiting factors. Different options of achieving large devices will be studied: changes in the device aspect ratio (number of fingers / unit finger width), changes in total device width, and the effectiveness of combining smaller devices (individual cells) to form a large total width device. E Frequency Capability We will study how the performance degrades with frequency, as we approach fmax. This study will not only show the range of frequency operation for the device, but also establish a relation between fmax and a frequency giving acceptable device performance (e.g. good gain and PAE). F Modeling In order to better understand the behavior of the observed power measurements, we will develop a simple, physics based device model. The model should capture the major dependencies observed in the data, and help in estimating the capabilities of the different CMOS technologies. 1.4. Outline We will start by a brief description of the experimental setup in Chapter 2. This includes the device layout, and the notations used to describe it, as well as the experimental setup for performing the different measurements. Chapter 3 will discuss the performance of 90 nm devices, and studies the impact of gate oxide and gate length scaling. It compares the performance of 90 nm, 130 nm and 250 nm technologies. Chapter 4 studies the performance of 65 nm devices, and also addresses the question of output power scaling. It shows the importance of device layout when scaling output power. A simple device model is developed in Chapter 5, which allows us to describe all the major trends seen in Chapters 3 and 4. We will combine all the results from the measurements and models in Chapter 6, which aims at illustrating the power and frequency limitations of CMOS technology for RF power. 4hh#' I- 10 r-- SAT 1: I -I ac a. 0.1 Deeply Scaled • CMOS "Performance : Space 0.01 -- , 0.1 i...... -, - m a i i..... I 1 H , , I I I I 10 1 'I 100 I I I I I I I1 I 1000 Frequency [GHz] Fig. 1.9 Illustration of how RF CMOS devices exhibit limitations of power and frequency. The goal of this thesis is to identify these limits, and study how they can be pushed out. Chapter 2. Experimental This chapter describes the devices and technologies used, as well as the types of experiments carried out on them. I will first describe the device technologies used and the types of devices. I will then broadly describe the three types of characterizations carried out on the devices: DC, Sparameter, and RFI power characterization. All measurements were performed on-wafer using standard RF probes (ground-signal-ground). 2.1. Technology Description The technology that has been studied in this work is a foundry 90 nm and 65 nm CMOS technology manufactured at IBM [6]. Table 2.1 shows the different devices fabricated. We will label these devices by their ITRS equivalent technology node: Lg,inn = 65 nm, 90 nm, 130 nm and 250 nm, with the understanding that the physical gate length will be below those values, particularly at the 90 nm node. The 65 nm wafer has devices with a nominal gate oxide thickness between 11 and 13 A and Lg,min = 65 nm (Vdd = 1.0 V). For the 90 nm wafer, in addition to the standard 90 nm NFET (Lg,min = 90 nm) with a nominal gate oxide thickness of 14 A (Vdd = 1.0 V), there is an intermediate oxide I/O FET (tox = 22 A, Lg,min = 130 nm) equivalent to a standard 130 nm node logic device (Vdd = 1.2 V), and a thick oxide I/O FET (tox = 51 A, Lg,nmn = 250 nm) equivalent to that of the 250 nm node (Vdd = 2.5 V). Depending on the process details, the additional mask count required to offer the I/O device options is between 1 and 3. The 90 nm devices were taped out May 2003, when the technology was ramped up for production. The 65 nm devices were taped out in August 2004, at which stage the technology was still undergoing development. The presence of the I/O devices allowed us to compare the performance of essentially four different CMOS technology nodes (65 nm, 90 nm, 0.13 ýtm and 0.25 Fam). The 0.13 [am medium oxide device, however, is not completely independently optimized. It shares some process steps with the nominal thin oxide 90 nm device. Therefore, it does not truly represent a 0.13 ý.m node, and its performance may differ. Table 2.1 highlights the key process parameters for these technologies: Parameter 65 nm 90 nm 130 Ipm 0.25 ýtm on 90 nm wafer on 90 nm wafer 0.25 pm ITRS Equivalent Node, Lgmin Physical Lphysical 65 nm 40 nm 90 nm 63 nm 0.13 tm 120 nm Oxide Thickness 11-13 A 14 A 22 A Nominal Voltage 1.0 V 1.0 V 1.2 V 240 nm 52 A 2.5 V Standard Cell (NFxWG,F) 64x12 pm 48x16 pm 34x16 pm 30x20 pm Table 2.1. Key device and process parametersfor thefour device types in this thesis 2.2. Device Layout We have fabricated devices with a variety of layouts providing a broad range of total device widths. The unit cell dimensions were given in Table 2.1. In general, the device layout consists of several multi-finger device cells in parallel. The total device width (WG,TOT) is the product of a number of parallel device cells (Nc), times the number of device fingers (NF) per cell, times the unit finger width (WG,F). We will label devices as WG,TOT = NC X NF x WG,F. Because the device fingers are very wide, electromigration concerns can arise due to the high current densities. A standard multi-finger device layout was modified to allow for better current flow in and out of the drain/source contacts, reducing the peak current within the device metallization layers. The layout uses a double-sided gate access to the device, as well as a double sided source access. The drain is routed out vertically. This metallization is done with 4 levels of metal. Thicker levels are then used to route the gate, source and drain to the RF pads. An illustration of the device layout is shown in Fig. 2.1. plane of symmetry Fig. 2.1 Vertical drain layout (source = red, drain = blue, gate=green).The structure is symmetric, and only half is shown. The symmetry plane is indicated. The layout design is patent pending. 2.3. DC and S-Parameter Characterization Measurements of the devices under DC conditions were performed at MIT using an HP4145B parameter analyzer. A sample Id-Vds sweep is shown in Fig. 2.2. This data was taken for the standard 90 nm device. S-parameter measurements were performed at IBM, on an automated probe station up to 80 GHz. From the S-parameter measurements, ft and fmax estimates were extracted. A typical S-parameter measurement result is shown in Fig. 2.3, for the 90 nm standard device. 400 300 200 100 0 0 1 2 3 Vds [V] Fig. 2.2 DC sweep for the standard90 nm device (1x48x16 tim), with Vgs from 0.4V to 1V in 0.1V steps. The DC measurements exhibit self-heating, which can be seen by the higher Ro as Vgs increases. Therefore, the Ro values seen in DC measurements for high Vgs and Id will differ slightly from those relevant to the RF operation of the device. 50x S12 D Fig. 2.3 S-parameters (0.5-50 GHz) for the standard 90 nm device (lx48x16 pIm), biased at Vds=lV, Vgs=0.58V. S11 and S22 are plotted on the left, and S21 and 50 times S12 are plotted on the right. 2.4. Power Characterization RF power characterization was carried out primarily at 8 GHz on a Maury load-pull system. We chose 8 GHz for the bulk of our work to explore the high-frequency potential of deeply scaled CMOS technologies. Measurements at higher frequencies also avoid device oscillations. The higher the gain, the more susceptible the device becomes to oscillations. If the impedances are tuned to maximize the gain at a high frequency, they will typically present a mismatch at lower frequencies. As the gain drops with frequency, this helps to flatten the gain vs. frequency behavior. This helps to avoid very high gain values at any frequency, and therefore reduces the risk of oscillations. The measurement theory and system design is described in more detail in Appendices A and B. The system creates a single RF frequency that is applied to the gate. Mechanical tuners set the source (gate side) and load (drain side) impedances to the device. The RF input and output power as well as the bias currents are measured, and allow computation of the relevant figures of merit (Id,PAE, Gain, Pout, Pin). Power measurements were optimized for peak PAE, by alternating source- and load-pull measurements backed off a few dB from the input power level giving peak PAE. This optimization was repeated for each change of device setup variables (for example, when measuring the same device at different values of Vdd). A typical power measurement sweep is shown in Fig. 2.4. As the device enters compression, the grain drops, while the output power saturates. The PAE is defined as PAE= out - Pin Pdc (Gain - 1) Pout Pd, Gain = Eq. 2.1 Pc where Pin and Pout is the RF power in and out of the device, and PDc is the DC power into the device. As the input power is increasing, the PAE will rise. However, in compression the gain drops as the output power is saturating. This will eventually result in the PAE beginning to drop, reaching a maximum PAE typically between 1-5 dB in compression. Often, we performed a series of power sweeps with one variable changing (e.g. frequency). The results of this series of sweeps were reported as the peak PAE and output power at peak PAE of each of the power sweeps. Separately, we also performed linearity measurements for the standard 90 nm device at 2.2 GHz on an ATN load pull system, at IBM. Throughout this thesis, we will be quoting the bias condition as a combination of Vds and Id. The value of Idis the DC drain current without the presence of input power. This current is achieved with a gate bias Vg. During the power measurement, this gate bias is held constant, rather than the actual Idvalue. In conduction type amplifiers, it is common to observe that as the input power is increased, the DC drain current measured into the device will also increase. The origin of the self biasing is the non-linearity that enters when the input signal swings below the device's threshold voltage [30]. While the average Vg remains at its bias value, swinging the input below Vt will not result in a negative drain current, but rather in zero current. This introduces the asymmetry. While a swing to higher Vg will result in higher drain current, swining Vg below Vt will not result in less (more negative) drain current. Therefore, the average DC current increases, despite the fact that the average voltage on the gate is unchanged. 70 - . 60 50 40 PAE 30 20 10 0 -10 -5 0 5 10 15 20 Pout [dBm] Fig. 2.4. A typical power sweep, showing the power added efficiency (PAE) and the transducergain, as a function of the measured output power. This measurement shows the 65 nm standard device (768 im width), at 4 GHz in class AB bias. Chapter 3 90 nm Technology The technology that has been studied in this work is a foundry 90 nm CMOS technology manufactured at I]BM [8]. In addition to the standard 90 nm NFET with a nominal gate oxide thickness of 14 A (Vdd = 1.0 V), there is an intermediate oxide I/O FET (tox = 22 A) equivalent to a standard 130 nmn node logic device (Vdd = 1.2 V), and a thick oxide I/O FET (tox = 51 A) equivalent to that of the 250 nm node (Vdd = 2.5 V). Additionally, devices with three different gate lengths are available. We label these by the half pitch of their equivalent technology node: 90 nm, 130 nm and 250 nm, with the understanding that the physical gate length will be below those values, particularly at the 90 nm node. Not all combinations of oxide thickness and gate length are implemented. Table 3.1 gives all available device options. Depending on the process details, the additional mask count required to offer these device options is between 1 and 3. We have fabricated devices with a variety of layouts providing a broad range of total device widths. The unit cell dimensions are given in Table 3.1. In general, the device layout consists of several multi-finger device cells in parallel. The total device width (WG,TOT) is the product of a number of parallel device cells (Nc), times the number of device fingers (NF) per cell, times the unit finger width (WG,F). We will label devices as WG,TOT = NC X NF X WG,F. Oxide Thickness thin (nominal) (14 A) medium (22 A) thick (51 A) 1.0 V 1.2 V 2.5 V Addl. Mask 0 1 to 3 1 to 3 48 x 16 jm n/a n/a L, = 90 nm 34 x 16 tm 34 x 16 tm n/a L,= 130 nm 30 x 20 /m 30 x 20 m 30 x 20 m L, = 250 nm Table 3.1: Gate length, oxide thickness and gate dimensionsfor the devices characterizedin this work. The device size refers to the number offingers x unit finger width. Sizes have been chosen to give identical drive currentfor the 90 nm thin-, 130 nm medium-, and 250 nm thick-oxide devices. Nominal Vdd 90 nm Standard Device Bias lx48x16 /m 8x48x16 im Vdd = 1 V Id= 25 mA 2.2 GHz 66 % 12.5 dBm 21.2 dB 25 dBm 14 % 6 dBm 11 dBm Vdd = V Id= 200 mA 2.2 GHz 59 % 20.2 dBm 14 dB 30 dBm 12 % 12 dBm 18 dBm Frequency Peak PAE Pout at peak PAE Small signal gain OIP3 PAE at IM 3=-35dBc Pot at IM 3=-35dBc Pout at 1 dB compression Table 3.2: Performance comparison of standard 90 nm device and a parallelcombination of 8 standarddevices. Impedances were optimizedfor the best linearity-efficiencytradeoff Power measurements were optimized for peak PAE, by alternating source- and load-pull measurements backed off a few dB from the input power level giving peak PAE. This optimization was repeated for each change of device setup variables (for example, when measuring the same device at different values of Vdd). Separately, we also performed linearity measurements at 2.2 GHz on an ATN load pull system. Additionally, S-parameter measurements were performed up to 80 GHz. 3.1. Results We will first discuss the RF power performance of the standard logic device (thin oxide, Lg= 9 0 nm), followed by a discussion of the performance impact of other device options that involve scaling the gate oxide thickness (tox) and gate length (Lg). In the following section we discuss the physical origin of the various results that have been obtained. 3.1.1. RF Power Performance of the 90 nm Logic Device The 2.2 GHz power performance of two standard digital 90 nm devices (biased at Vdd=1 V) with different total[ gate widths is shown in Fig. 3.1, and summarized in Table 3.2. We compare the performance of a single cell device (1x48x16 Im) to that of another device consisting of 70 0 60 -20 ,.50 -40 < 40 M 30 -60 20 -80 10 -100 0 -30 -20 -10 0 10 20 30 Pout [dBm] Fig. 3.1 Power performance at 2.2 GHz of a thin-oxide 90 nm device (1x48x16 /Am) and a device array (8x48x16 iLm), both biased at Vdd = 1 V and Id = 26 mA/mm. Impedances were optimized for the best linearity-efficiency tradeoff. eight of these cells wired in parallel (8x48x16 Am), giving an eightfold increase in WG,TOT. For the 1x48x16 Am device, a peak PAE in class AB operation of about 66% is obtained at an output power of 12.7 dBm (18.6 mW, or 24.2 mW/mm). The 8x48x16 Am device achieved a peak PAE of 59% at an output power of 20.2 dBm (104 mW, or 17.0 mW/mm), indicating a good scaling of Pot with cell count. A slight decrease in PAE, power gain and output power density can be seen. This indicates that an upper limit on power scaling exists. The PAE at linearity (in terms of IM 3) of -35 dBc is 14% and 12% for the lx48x16 Am device and the 8x48x16 Am device, respectively. To put these results in context, a WCDMA power amplifier driver requires about 14 dBm at 1 dB compression power, with an OIP 3 of 24 dBm [9]. From Table 3.2 we see that the 8x48x16 Am device exceeds these requirements by a safe margin. The results of Table 3.2 suggest that 90 nm CMOS is quite suitable for a wide range of integrated PA D - IV0 'out Air= 80 10 m .0 0 w 60 c 40 - - 21. -5 ," 0 20 1x4 8x16) V dd=1 V, Id= 2 6 rA/mm 0 - I I I 0 5 10 15 -0 20 Frequency [GHz] Fig. 3.2 Peak PAE and output power at peak PAE as a function of frequency for the standard 90 nm device (1x48x16 ALm). All measurement points were taken at Vdd = V, Id = 26 mA/mm, and impedances were reoptimized at each frequency. The output power is roughly independent of frequency, while the peak PAE is dropping as the frequency increases. applications, such as WLANs, bluetooth and cellular PA drivers, giving acceptable efficiency under linear conditions. Measurements at 8 GHz for the standard 90 nm thin-oxide device at Vdd = 1 were also performed. The standard device (1x48x16 rim) achieved a peak PAE of 56.8% at 13.4 dBm (21.9 mW, or 28.5 mW/mm) output power with an associated gain of 14.6 dB. These results at 8 GHz are slightly better than the performance reported at 2.2 GHz above. This is due to using a different measurement system that is able to present a better source and load match. We have also measured the power performance of the standard device (1x48x16 rpm) as a function of frequency. The results are shown in Fig. 3.2. The output power is constant with frequency over a broad range, while the peak PAE is dropping with frequency. Operating the device at higher frequencies reduces the power gain and results in lower peak PAE. Across-wafer uniformity measurements of 18 devices were also performed. These measurements were carried out with constant impedances, input power and bias points (Vdd = 1 V,ld = 26 mA/mm) at 8 GHz. The data includes both the probe contact uncertainty as well as the device variation across the wafer. The device geometry used was lx96x8 Jim. Excellent uniformity was obtained across the entire 8" wafer, with a measured PAE of 59.3%±0.6%, and an output power of 14.2 dBm ± 0.1 dB, as well as a small signal gain of 16.3 dB ± 0.3 dB. This difference in performance is due to the different device layout (1x96x8 Axm vs. lx48x16 Am). RF Power Performance of Other Device Options 3.1.2. We have studied the RF power suitability of the different device options as a function of the drain bias, Vdd. This provides a fair performance comparison as the different devices will show optimal operation at different voltages. Fig. 3.3 compares in detail the peak PAE power performance of the thin oxide devices for the three different gate lengths, as a function of Vdd at 8 GHz (Id= 26 mA/mm). The graph shows the small signal gain, the peak PAE and the corresponding output power density normalized to a 1 mm device width. At any one value of Vdd, there is not much difference between the 90 nm and the 130 nm devices, while the 250 nm device gives much lower performance. As Vdd increases, 7u - IUUV -- --- o 60- 0) 50 - - - Lg=90nm Lg=130nm -Lg=250nm PAE 100 E 40 a Em S30- -100.C E 30 - < C ) -vdd7 20 1 freq = 8 GHz 10 - thin oxide 0 Id = 26 mA/mmrr5:- I 0.1 1 ifW I 1.0 Vdd [V] i i i 1 I 10.0 Fig. 3.3 Gain, PAE and output power density at peak PAE point as a function of Vdd for thin oxide devices of three different gate lengths at 8 GHz. Impedances and input power drive were reoptimized at each Vddfor each device. '· IVVV Iu -c--Thin Oxide " C ," 60- -- - Medium Oxide Thick( 50- 100 E MAO 2 .Vdd - S20 - 10 freq = 8 GHz C. Id = Z6 mA/mm lx30x20 pm L.g = 250 nm 0 i i i 0.1 , i l ii 1.0 l , 1 i i 10.0 Vdd [VI Fig. 3.4 Gain, PAE and output power density at peak PAE as a function of Vddfor 250 nm devices with three different gate oxide thicknesses at 8 GHz. Impedances and input power drive were reoptimized at each Vddfor each device. all devices show a higher output power density. The dependence is quadratic, and agrees with the behavior seen in chapter 1 (Fig. 1.3). The gain at peak PAE changes only slightly with Vdd and is not very different from one device to the other. For the 90 and 130 nm devices, the peak PAE is rather flat across a broad range of Vdd. The 250 nm gate length devices, however, not only show a much lower peak PAE, but the PAE starts to rapidly drop as Vdd is reduced. The performance of the Lg=250 nm devices is limited by the low voltage capability of the thin oxide. We discuss the physics of this in section 4. Fig. 3.4 shows the impact of gate oxide thickness, keeping Lg constant at 250 nm. Decreasing the oxide thickness without changing the gate length results in lower gain and lower PAE. However, the power density for a given Vdd is increasing slightly. The reason for this is discussed in Section 4. ---- Lg=90nm, Thin Ox 70 - ---- Lg=130nm, Med Ox 1000 -L-- l=250nm, Thick Ox 60 50 - 100 i 40 - aC) 30 - ,",-Vdd 2 10 2010 - freq = 8 GHZ h~-~·~ac~brhca~-~ .uCIII I 1 · · I·I r a- Id= 26 mA/mm I 1 0.1 · 1 I · 1I r 1.0 Vdd [V] 1 1 10.0 Fig. 3.5 Gain, PAE and output power density at peak PAE point as a function of Vdd when scaling both Lg and gate oxide thickness at 8 GHz. Impedances and input power drive were re-optimized at each Vddfor each device. S- Vdd = 1V Vdd = 2.5 V 60 90 nrr thin o: 50 [fllUK OXIUU 40 - 30 10 100 1000 Output Power Density [mW/mm] Fig. 3.6 Comparisonof the 90 nm thin and 250 nm thick oxide devices. Both devices can achieve similar peak PAE levels, though at different operatingvoltages andpower densities, as indicated. Finally, Fig. 3.5 shows the RF power performance when we both scale oxide thickness and Lg. This is equivalent to comparing three generations of CMOS technology. For high enough and voltage, the 250 nm I/O devices and the 90 nm devices reaches identical levels of peak PAE gain. The advantage of the 250 nm device is that its thicker oxide allows higher overall power levels. At Vdd = 2.5V, the device achieves a peak PAE of 59.5 %, at a power of 18.7 dBm (74.2 mW or 124 mW/mm). The performance obtained in these 250 nm I/0 devices is better than reports available in the literature for nominal 250 nm node logic CMOS devices [10]. When comparing all of the devices, two devices clearly emerge as the best performing ones: 90 nm standard device and 250 nm logic device. Fig. 3.6 shows a comparison of these two devices, using power and PAE as our variables. The optimum device choice for RF power applications Vu 0nn -·- - AA- Lg= 90 nm E Lg = 130 nm E Lg =250 nm 50 N I . 40 - 150 A' Ca ; :··· ·:::.. o 304- ~:·:: -r_, 20a- - 100 0- E: 10 10- : : · `:·::· :: ': -F v - thin a-50 medium thick Gate Oxide Thickness Fig. 3.7 Comparison of fma for the devices in Table 1 operating at Vnominal. Small width devices (right axis) with their pad parasitics de-embedded offer a peak fma of 100 to 200 GHz. The figure shows the strong impact of the device layout on the performance. The fm of the power devices (left axis) used in this paper are not de-embedded, and show a much lower fm due to the low bias (Id = 26 mA/mm) and the parasitics associatedwith the pads, the large number of fingers and their width (e.g. R, and Cgd). depends on the system design constraints. If an optimum voltage can be used, Fig. 3.6 suggests that the thick oxide 250 nm device operating at higher voltages delivers a higher output power density than any of the thin-oxide devices, while operating with identical peak efficiency. For low voltage operation, the thin oxide devices are preferable because their lower knee voltage leads to a higher peak PAE, as seen in Figs. 3.4 and 3.5. 3.2. Discussion In order to understand the RF power performance, it is very useful to study the behavior of fmax. In our research and in other research it is shown that fmax provides a very strong correlation with peak PAE [ 11]. In order to understand the physical origin of the results shown in Figs. 3.33.5, we have measured ft and fmax for all these devices at their nominal Vdd (given in Table 3.1) and at IL= 26 mA/mm. The fmax results are graphed in Figs. 3.7. The fmax values are relatively low because the current bias point is not optimal for fmax, and because additional parasitics such as Rg and back-end parasitics penalize very wide devices. The fmax data is not de-embedded, because pad and wiring parasitics are an unavoidable part of the PA. For comparison, the fmax of typical small analog devices in this process is also shown in this graph. These devices have the pad and wiring parasitics de-embedded. The standard 90 nm logic device attains an optimum fmax of about 200 GHz, which is characteristics of state-of-the-art 90 nm technology [12-14]. In contrast, the power device at the optimum power point only attains 32 GHz. This data helps us to understand many aspects of the results shown in the previous section. An approximate formula [15] for fmax using the device's ft as well as gate resistance (Rg), channel resistance (Ri), output conductance (gd) and gate-drain capacitance (Cgd) is given by f 3.1 ft = fn 2(Rg + R)(gd +24fCgd Fig. 3.3 shows the impact of Lg on performance. The Lg = 90 and 130 nm devices show about the same performance, while Lg = 250 nm gives a much worse performance due to the low fmax (gain at 8 GHz). Fig. 3.7 shows that the 90 and 130 nm thin oxide power devices have an identical fmax. These two devices show the same fmax because as Lg is increased, the effects of lower parasitic gate resistance (Rg) can cancel out with the lower ft (Equation 1). The 250 nm thin oxide device, however, shows a much lower fmax. This is equivalent to lower power gain and subsequently leads to a lower peak PAE, as observed in Fig. 3.3'. The results of this comparison indicate that the optimum device Lg for power performance may not necessarily be the minimum Lg. As Lg is increased, the device's ft decreases. But at the same time, the parasitic gate resistance is reduced. Thus, a sweet-spot can exist where fmax reaches a maximum in the tradeoff between ft and Rg. For the 90 nm process, as Fig. 3.3 shows, that optimum gate length seems to be somewhere between the 90 nm and 130 nm nominal gate lengths. 250 nm, thick oxide 90 nm, thin oxide OW i 00 e-•5 400 400 300 300 S200 200 100 100 0 - 0 AX/ 0 0 0 2 1 Vds V] 3 0 2 1 3 Vds [V] Fig. 3.8: Load-line comparisonfor the 90 nm thin and 250 nm thick oxide devices. MeasuredDC I-V characteristicsof the two devices are shown, with the approximate Vds,sat indicated. The 90 nm device's Vds,sat is lower than that of the 250 nm device. Load-lines, based on the impedances used in the power measurements of Fig. 3.5, are also shown at Vdd = IV and 2.5V. Self-heating is present in the DC characteristicsof both devices. The impact of the oxide thickness in power performance (Fig. 3.4) is not as dramatic as that of Lg in Fig. 3.3. The use of a thinner gate oxide will reduce the voltage capability of the device. Therefore, while the performance is not dramatically different at a constant Vdd the thin oxide devices cannot be operated at the same high Vdd values. The thick oxide device can therefore achieve the to best overall peak PAE and output power density due to higher Vdd. Decreasing the gate oxide thickness without changing anything lowers fmax (Fig. 3.7). This explains that the thick oxide device shows a slightly better peak PAE at any given Vdd. The comparison of the three device technologies (90 nm thin-, 130 nm medium- and 250 nm thick-oxide) was shown in Fig. 3.5. If we compare the devices at a fixed Vdd, we notice that the power density of the 250 nm thick oxide device is lowest. Further, the peak PAE for the 90 nm thin oxide device is independent of Vdd down to about 0.5 V, while the 250 nm thick oxide device's PAE begins to drop below Vdd = 1.5 V. This difference in behavior can be explained by plotting the load lines at the peak PAE point, as shown in Fig. 3.8 on top of the output characteristics. The load lines were calculated from the load-impedances, and superimposed on measured DC characteristics. In this figure, we compare the 90 nm thin oxide to the 250 nm thick oxide devices. The thin oxide device has a lower Vds,sat. Therefore, if both devices are operated at Vdd=lV, the load-line for the 250 nm thick oxide device intercepts Vds,sat earlier. This results in earlier and softer compression, and therefore lower output power density and PAE. If, on the other hand, Vdd for the 250 nm thick oxide is raised to its nominal voltage of 2.5 V, the relative impact of Vds,sat on the load-line is reduced. The device will now compress later and sharper than at Vdd=IV, allowing it to achieve PAE levels comparable to the 90 nm thin oxide device. The output power density will be higher than that of the 90 nm thin oxide device at Vdd=lV, because Vdd is 2.5-times higher. This can be seen in Fig. 3.5. 3.3. Conclusions We have studied the RF power suitability of the various device options offered in a typical 90 nm digital CMOS foundry process. We have demonstrated excellent performance at 2.2 and 8 GHz that allows implementation of integrated PAs for several wireless consumer applications. We have found that for low voltages the 90 nm nominal devices offer the best output power and peak PAE, but 130 nm thin-oxide devices closely match it. For higher voltage operation, the 250 nm thick-oxide devices offer the highest output power and peak PAE. Chapter 4. 65 nm Technology The 65 nm devices studied in this chapter were fabricated in near-production 65 nm CMOS [16] technology flrom IBM. We compare the RF power performance of this technology with similar devices made in IBM's 90 nm process and 0.25 tm I/O devices fabricated in the same 90 nm technology [1], which were introduced in Chapter 3. A summary of the devices is given in Table 4.1. We have fabricated devices with a variety of layouts providing a broad range of total device widths. Technology Lg (nm) Tox (A) 65 nm 65 11-13 Vdd (V) 1.0 90 nm 90 14 1.0 90 nm 250 51 2.5 Table 4.1: Overview of the devices used in this chapter,fabricatedin either 65 or 90 nm technologies. The gate length refers to the equivalentpolysilicon halfpitch dimension. 4.1. Experiments Fig. 4.1 shows the RF power performance of two standard 65 nm digital devices at 4 GHz and a bias of Vdd = 1 V with Idq = 26 mA/mm. Both devices have the same total gate width of 768 Am but different layout configurations. The first device is made up of 12 parallel cells consisting each of 16 fingers with 4 Am unit finger width (we will use the notation: 12x16x4 Atm). This device shows an output power of 13.8 dBm (31.2 mW/mm) at a peak PAE of 65.7% when biased at the nominal voltage of 1 V. The small-signal gain is 17.1 dB. The second device uses a single cell (1x64x12/zm). Its peak PAE is 62% with 13.8 dBm of Pout and a small signal gain of 14 dB. The lower performance of this second device suggests that the optimum gate finger width for this technology is closer to 4 jtm rather than 12 .Lm. While the PAE results are high, Pout per unit gate width is smaller than for lower-integrationlevel, higher-voltage technologies [17-21], which can exceed power densities of 1000 mW/mm. Many applications, such as PA drivers or wireless LAN PA's require absolute power levels in excess of 20 dBm. In principle, there are three ways to accomplish this: increasing Vdd, increasing bias current (Idq), or increasing overall device width (WG,TOT). We have examined all three approaches for the 65 nm CMOS node. /U 60 2. 50 c 40 o 30 < 20 10 0 -10 -5 0 5 10 15 20 Pout [dBm] Fig. 4.1 Power characteristicsof two 65 nm devices with a total gate width of 768 Jim at 4 GHz. The 12 cell device delivers a peak PAE of 65.7% at a power of 13.8 dBm (31.2 mW/mm) at a voltage of I V. ou - Fren=rS GHz~V..=1 V. L=2A mAlmm 12x16Cx~pm *1l*** 50 -20 0~~ 40 - - W. (5 -10 I, I < -20 - a. -Vdd J E C, 10 - 00.01 -10 I I I I I I I 0.1 I Vdd IT i 7I I i 1 I I .I I --10 10 Fig. 4.2 Power performancefor the 12x16x4 AIm, 65 nm device at 8 GHz vs. Vdd. Foreach Vdd, the device is tuned to peak PAE. The gain is the small signal gain. Fig. 4.2 shows the RF power performance of the 65 nm logic device as a function of Vdd at 8 GHz. The device is biased at a constant current and tuned to the peak PAE point for each change in Vdd. We find that Pout increases quadratically with Vdd, as expected from simple theory. PAE exhibits a broad plateau with a best value of 53 %being obtained at around the nominal voltage of 1 V. PAE is found to drop for very low values of Vdd as well as for Vdd beyond 1 V. At low voltages, PAE is limited by Vds,sat and at high voltage by impact ionization. This can be seen in the output characteristics of this device in Fig. 4.3. Since reliability is also expected to be rapidly compromised for Vdd>l V [1], 1 V represents a reasonable limit for Vdd for this technology in this application. Fig. 4.4 shows the power performance as a function of the bias drain current, Idq, at 8 GHz and at Vdd=l V. The figures shows PAE, Pout, the small signal gain, as well as actual DC drain ,' ,~ V.3I 0.2 r-o 0.1 0 0.2 0 0.4 0.6 0.8 1 1.2 Vdd [V] Fig. 4.3 V steps. For Output characteristicsof the 12x16x4 Itm, 65 nm device for Vg = 0.2 to 1 V in 0.2 small values of Vgs and high values of Vds, impact ionization effect can be seen. Overlapping the device characteristicsis the load line correspondingto a biaspoint of 1 V. V r% A < E 50 25 20 0f 45 U, 40 a t- 15 35 10 a030 E 5 25 1 10 100 IDQ [mA] Fig. 4.4 Performanceof a 65 nm device (12x16x4 jtm) as a function of bias current. Idq is the quiescent drain current (no Pin), and Id is the drain currentat peak PAE. The device is self biasing and the peak PAE, Pout and Gain do not change much with bias current. current flowing through the device at the peak PAE point (Idq refers to the DC current without RF signal being applied). Interestingly, the RF power performance is broadly insensitive to the actual value of Idq. The peak PAE only varies from 50 % to 52 %, and Pout from 20 to 26 mW, over an 1 x increase in Idq. This is because of the high degree of device self-biasing that is taking place. While Idq increases by more than a decade in this figure, the actual DC current at the peak PAE point only increases by about 30 %, from 36 mA to 47 mA. The third and potentially most effective way to attain more power is through gate width scaling. Fig 4.5 shows the impact of WG,TOT scaling on the RF power performance. In this figure, WG,TOT is varied by connecting increasing numbers (Nc) of fixed-size 64x12 tim cells in parallel. As WG,TOT increases, Pout increases initially but then saturates to a maximum power - - S- nf\ V 50 - 150 40 - eq = 2.2 GHz w 30 - - 100 nm .. 20 - - LE - - 50 10 - Freq = 8 GHz Vdd = 1V, Id= 26 mA/mm 00 O - 11.11.<. 1 4 5 6 2 3 Total Device Width [mm] 7 8 Fig. 4.5 Powerperformance as afunction of device size for 65 nm devices (1, 2, 3, 4 and 6 parallelcells of 64x12 jim each) and 90 nm devices (1, 2, 4 and 8 cells of 48x16 Jim). As the number of cells and thus device width increasesfor the 65 nm devices, PAE drops while Po,t saturates. The behavior of 90 nm is very different as Pot continues to increase with device size. Given the independence of power on frequency over the range that is studied in this work (see 4.), we have includeda data pointfor a very wide 90 nm device (6.1 mm) that was measuredat 2.2 GHz. A*AA .Ik Fig. 4.6 (a, left) Simplified model schematic to create a multi-cell device from a number of single cell devices (b, right). The intrinsic device cells of the single cell and the multi-cell device are identical. Adding parallel single cell devices requires parasitic resistances to match the increased wire resistance of the multi-cell device's backend. The model parameters were extracted using S-parametermeasurementsfrom a single cell and the multi-cell device. E 7. Frequency [Hz] 1( Frequency [Hz] Frequency [Hz] Fig. 4.7 Y11 comparison of a two-cell device (2x48x16 Aim) with the model. When using two parallel 1x48x16 /m device measurements, the match with the two-cell device is mediocre. Adding parasiticresistances on the source, gate and drain, as shown in Fig. 4.6b, results in a much better match with the two-cell measurements. The other Y-parameters exhibit a similar picture and are omitted to preserve clarity in the graph. level of about 49 mW. PAE drops correspondingly. This is an anomalous behavior which deserved further study. Regardless of its origin, the power level of about 50 mW that was attained is nevertheless sufficient to enable lower-power system-on-chip applications such as Bluetooth Class 2 and 3 [22]. We have obtained excellent efficiencies and gains at very small voltages and relatively high frequencies. However, power scaling through device size appears to be limited to about 50 mW. The next section will discuss the cause we believe is responsible for this phenomenon, and suggest ways to mitigate it. 4.2. Discussion In order to understand the origin of this maximum power limit for large multi-cell devices, we have carried out S--parameter measurements and extracted a small signal equivalent circuit model for these devices as a function of the number of cells that are placed in parallel. We have measured the S-and Y-parameters for a single cell device. Ideally, a multi-cell (Nc) device should exhibit the same Y-parameters as Nc individual single-cell devices wired in parallel (Fig. 4.6a). However, when we do this in the simulation environment, we find significant discrepancies (Fig. 4.7). These differences between the multi-cell device and Nc single cells in parallel must arise from differences in the backend wiring parasitics, as the intrinsic device is identical. Fig 4.7 also shows the performance of a revised multi-cell model (Fig 4.6b) that take into account the added parasitics in a multi-cell device. Parasitic resistances in the gate, drain, and source of each individual cell were required. We optimized these resistances for a close C 4u En YE -0 a:> 0 1 2 3 4 5 6 7 Number of Cells Fig. 4.8 Small signal extracted parasitic scaling resistances (Rg and Rs+Rd). To construct a multi-cell device from single-cell devices, these resistance must be added into the single cell unit (NFx WF = 48x12 tim = 768 lim) as shown in Fig. 4.6b.. match with the measured Y-parameters of the multi-cell device. We are assuming that the added source and drain resistances are the same, because their impact on the Y-parameters is very similar, which makes it hard for the optimizer to distinguish them. The presence of the resistances helps to improve the Y-parameter match, as we saw in Fig. 4.7. Fig 4.8 plots the evolution of the additional back-end parasitic resistance associated with the gate and the source plus drain. This resistance is required in order to match the multi-cell Yparameter measurement with an increasing number of single cells, placed in parallel. As the number of cells increases, the parasitic BEOL resistances per unit cell increase as well. This is because, due to pads, backend wiring, and geometry constraints, not all metal wire sections scale in proportion with the number of cells. The increased resistance results in source degeneration and resistive power loss, and therefore lower PAE, gain and Pout. Cr\ OU - "" /uu ideal 40- 150 100 - a. -x 20010 ideal CL 50 - - ai 0 l l l 8 10 2 4 6 Total Device Width [mm] s 12 I 0 0 I I I I I 2 4 6 8 10 12 Total Device Width [mm] Fig. 4.9 Model of the peak PAE (left) and Pout performance (right) with scaling, matched to the 768 1lm device, and using the resistancesfrom Fig. 4.8. The graph also shows the ideal scaling behavior, as well as that of reducing the extracted resistanceby 50%. To prove this, Fig. 4.9 shows the RF power performance prediction of a simple model, based on the same resistances shown in Fig. 4.8, as well as half of that resistance. These simulations have also been carried out in ADS. The model assumes that the performance of each unit cell is identical, except that the parasitic drain and source resistances (Rparasitic) reduce the fraction of DC power (PDc) that is converted into RF power (Pout) in the device. The DC power will scale with the number of cells: PDoc, = Nc IDOVdd Eq. 4.1 where Nc is the number of parallel device cells, and IDO is the drain current for a single cell device. Because of the parasitic resistance, some of this power will be lost, and not all the DC power reaches the intrinsic device. If we denote Ppa. as the DC power lost in the resistor, we have: Ppara = c(Nc ID )2RN Eq. 4.2 where c is a constant depending on the waveform, typically between 1 and 2, RN is the additional parasitic resistance for the Nc cell device, PAEo is the peak PAE for the single cell device, and PDCN is the total DC power for the Nc cell device. The RF output power will then depend on the amount of DC power into the intrinsic device, as well as the efficiency of the device to convert it into RF power: Eq. 4.3 Pout,N = (PDC,N - Ppr )PAEo where Pout,N is the output power for the Nc cell device. Finally, the NC cell PAE is computed from the RF power and overall DC power: PAE PouN Eq. 4.4 out PDC,N where PAE is the power added efficiency for the Nc cell device. For a fixed PDc, the loss in Pout results in a lower PAE. The power loss is proportional to the RN, and Pout is a linear function of RN. The model is optimistic, as it does not take into account source degeneration, nor the nonideal shape of the output waveform. It also omits the self-biasing aspects, which will impact the load-line. In spite of these shortcomings, the model captures the relevant scaling behavior and shows the sensitivity of performance to the resistance. Fig. 4.10 shows the behavior of the peak PAE and Pout as a function of device width. We can combine the equations 1-4 to obtain more compact expressions for PAE and Pout: PAE = ( clR Nvdd NC PAE o Pout,N =(Nc IDooV, -c(Nc Io)2 R )PAEo Eq. 4.5 Eq. 4.6 The scaling behavior with Nc is now easier to analyze. Eq. 4.5 shows that PAE falls linearly with Nc, as seen 'in Fig. 4.6. Eq. 4.6 shows that the output power follows a parabolic shape. Therefore, a maximum for Nc can be found, which is given by: N Eq. 4.7 Vdd 2clIoRN PAENsat = IPAEo 2 S2 Psat 4cRN Eq. 4.8 PAE Using this, we can compute the output power and PAE at the maximum power level. These values are shown in Eq. 4.8. It is interesting to note that when the peak output power is reached, the model predicts the PAE to have dropped to 1/2 of its original (Nc small) value. This simple model shows the critical impact of the interconnect resistance on the RF output power that can be achieved. Carefull layout optimization and use of existing 4x-thickness upper metal levels for cell wiring should ameliorate this deleterious effect and enable the 65 nm node to scale the power level in a similar fashion to the 90 nm node, at equal supply voltage. The increased inductance or capacitance of the additional metal can, in principle, be resonated out with the source and load impedances. The model also suggests that an improvement in the efficiency will allow the designer to wire more devices in parallel and achieve a higher overall power level before PAE drops below the desired minimum. In Eq. 4.6, the output power is roughly proportional to the single cell PAE (PAEo). 60 50- Freq = 8 GHz Vdd=l V, ld= 2 6 mA/mm = 1 12xl 6x4pm U 1 cell 64x12pr cell '40- ells 9 30- s 6 cells n 20- 10 0 20 40 30 50 fmax [GHz] Fig. 4.10 Correlationof peak PAE with fm (extracted at V8s = 0.4 V and Vds=l V) for the 5 different 65 nm device sizes (each cell is a 65 nm device of 64x12 im), as well as the optimized device layout (12 cells 16x4 tm devices) I,,' V0 - standard cell e4'f~jrf'ftC _9 40N I 03 20 m 1xNFxWC Ar_ __ • • 65nm: Vdd=lV, Vg=U.tj V 90nm: Vdd=lV, Vg=0. 7 5 V I I I I I 65 nm I i I I 10 I II100 100 Unit Finger Width (WG,F) [um] Fig. 4.11 fmax vs. unit finger width for a 1 cell device, keeping overall width the same (NF x WG,F = 768 jpm). Large finger width hurts fm because of Rg. Narrow finger width hurts because of distributedeffects involving the large number of device fingers that must be parallelized We are interested in studying in detail the dependence of PAE on various layout decisions. Doing this with sets of power measurements is difficult, because time constraints would limit the number of bias points and devices that could be studied. Instead, we are using fmax as a predictor for PAE, which has a high correlation with the measured peak PAE. This can be seen in Fig. 4.10, which compares the peak PAE and fmax for devices of various total width. The peak PAE performance and f.x are highly correlated. If fmx, is much larger than the frequency of operation, the peak PAE eventually levels off to a rather high value. Because fmax correlates so tightly with PAE, we will next discuss strategies to improve fmax (and therefore power performance) through layout optimization. Fig. 4.11 and Fig. 4.12 show the impact of device layout on fmax for a device Total width: 768 um IE2I V E50 U- C40 ' 20 E 100 0 1 10 100 1000 Number of Cells (Nc) Fig. 4.12 Breaking the 65 nm device up into multiple cells in parallel, both the number offingers and their unit width are scaled down at the same time. This method can achieve higherfax values (51.4 GHz vs. 43.7 GHz) at identical bias conditions. However, too many cells in parallel hurt performance due to the excessive wiring required,as seen by the drop in fmx as the number of cells increases beyond 50. with a fixed WG,TOT = 768 Am. While the fa for small logic devices is well above 200 GHz [23,24], it is substantially lower for the large power devices. This is very similar to what we observed in the 90 nm node [1]. Fig. 4.11 shows the impact of the device aspect ratio of a single cell device on fma. The abscissa plots the unit finger width. As this increases, the number of fingers that must be placed in parallel to obtain a given total width has to decrease. An optimum value of unit finger width is obtained at about 6 um. For narrow fingers, the number of fingers that must be parallelized is large and this introduces increasing distributive losses. For wider fingers than the optimum, the finger gate resistance increases and this degrades fmax. In Fig. 4.12 we show the impact of device cell size, as we break a cell into many sub-cells in parallel. We can see a large improvement in fmax when going from a single cell with many wide fingers to an array of many cells with shorter and fewer fingers per cell. Throughout this 4 D Iuu -- r out 80 Qn nm m 60 40 - a. 20 0 0 6Vdd 0 -5 a- Vdd--- 0 Id= 26 mA/mm a a : 5 10 15 20 Frequency [GHz] Fig. 4.13 Performance comparison of the 90 nm and 65 nm devices as a function of frequency. Both devices have a total width of 768 ALm (65 nm: 64x12 Alm and 90 nm: 48x16 Alm). The output power is independent offrequency, and essentially identicalfor both devices, while the peak PAE for this 65 nm device is slightly lower. experiment, the total device size WG,TOT (N x NF x WG,F) remains constant. The data of Fig. 4.12 explains the results of Fig. 4.1: a large improvement in fma can be obtained by connecting many smaller devices (cells) in parallel. Fig. 4.12 suggests that the optimum cell size is 16x4 Itm, with 12 cells in parallel achieving a total width of 768 jm. 4.3. Technology Scaling: 65, 90 and 250 nm We now compare the RF performance of three different generations of CMOS technologies, combining the 65 nm results with those obtained on a 90 nm wafer [25] containing 90 and 250 nm devices. The longer 250 nm devices on the 90 nm wafer are used to drive input/output pads and approximate both the structure and performance of its corresponding CMOS generation. The device geometries are 1x64x12 /.tm, 1x48x16 gm, and 1x30x20 Jm for the 65 nm, 90 nm and 250 nm devices, respectively. The change in finger width is necessary to compensate for the increasing gate resistance as Lg is decreased. Fig. 4.13 shows the peak PAE and Pout performance as a function of frequency, for single cell nominal 90 nm and 65 nm devices. While the peak PAE performance of both devices drops rapidly with frequency, the Pout at the peak PAE point is largely unchanged. This is expected. There is no substantial difference in behavior between the 90 and 65 nm devices, though the 65 nm device has a slightly lower peak PAE across all frequencies. The similarities of the two technologies can also be seen in Fig. 4.11, where the two standard cells (1x64x12 utm and lx48x16 /m) achieve roughly identical fx,, figures. Fig. 4.14 shows the Vdd behavior of the three technologies. This graphs shows the peak PAE and output power density vs. Vdd at 8 GHz for devices with an identical WG,TOT. We note that, at -L/U Lg=65nm (1x64x12 um) '- m Ia 4-AC-4 60 ,- 50 100 LU 40 0 o E -30 a) 20 10 0- 10 0 0.1 1.0 10.0 Vdd [V] Fig. 4.14 Peak PAE and output power density at 8 GHz as a function of Vdd for three different CMOS technologies. their respective nominal voltage of Vdd = 1 V, the 65 nm devices have a somewhat lower performance than the 90 nm and the 250 nm device on the 90 nm wafer, likely due to the higher interconnect resistance described above. For high values of Vdd, the 65 nm device exhibits slightly inferior performance when compared to the 90 nm and 250 nm devices. However, up to about Vdd = 1 V, the 65 nm device has similar performance as the 90 nm device. The 250 nm device yields a lower power density at a fixed Vdd, but because it is able to operate at a much higher Vdd, the maximum power density that it can achieve is the largest of all three technologies. While the device performance of the 65 nm and 90 nm devices is very similar with respect to Vdd and frequency, a significant difference in performance was seen when looking at the device width scalability (Fig. 4.5). Both technologies show a decreasing PAE with increasing device width. However, as noted earlier, the power level of the 65 nm devices saturates while that of the 90 nm devices continues to rise as the number of cells in parallel increases. While some of this arises from back-end scaling, the largest difference between the 90 nm and 65 nm experiments is a change in the metal routing used for the multi-cell device layout. It is important to note that the use of the thicker back-end metal process options, wider metallization and a layout that scales well with the number of cells, should allow a more mature 65 nm technology to perform on par with the 90 nm technology. Optimizing the source and load impedances will resonate out the additional parasitic capacitances and inductances. For low power levels (below 16 dBm or 40 mW), the 65 nm technology CMOS node offers excellent efficiency over a broad frequency range (up to about 10 GHz). Its RF power performance approaches that of 90 nm devices both in peak PAE and output power density. The relatively high back-end resistance of the BEOL of 65 nm CMOS leads to difficulty in scaling output power by wiring many fingers and cells in parallel. A maximum power level of 47 mW at Vdd = 1 V has been obtained. Simulations indicate that further optimization of device layout, including the use of stacked or thicker upper metal levels should mitigate the negative effects of BEOL resistance scaling. Through this, the 65 nm node can provide efficient integrated power amplifier functionality for many applications, even in a deeply-scaled logic CMOS technology without costly PA.-specific adders. Chapter 5. Power Device Model In chapters 3 and 4 we discussed the RF power performance for many different device geometries and technologies (65 nm, 90 nm, 130 nm and 250 nm). We also demonstrated how the power performance varied as a function of bias and frequency. This chapter introduces a simple device model that provides understanding of the different dependencies that have been observed. The goal is not to create a highly accurate but complex model - but rather to have a very simple, physics-based model that is able to predict the correct trends in the power performance and that helps us to chart the capabilities of future CMOS generations. Fig. 5.1 shows the measured DC characteristics of the standard 90 nm device (1x48x16 .tm), as well as the sketch of a typical load-line in compression. This figure shows the regions of the IV characteristics one needs to focus on when developing a device model for RF power. We can see from the figure that the load-line sweeps through the sub-threshold region at high Vds>>Vdd. For Vds<<Vdd, the device enters the linear region (Vds<Vds,sat) and the load-line will extend until the on-resistance (Ron) limits the drain current. The model must be able to capture these key regions of the DC characteristics. Other concerns in power operation are the maximum Vds that the device can operate: at (breakdown voltage), and the RF loss on the gate and drain as a function of frequency. The model described below captures all these critical parameters. We will use this model to simulate: device behavior in Agilent ADS, using harmonic balance simulations. This helps us to get a more accurate picture into why performance deteriorates with frequency, and how the output power scaling seen in chapter 4 exhibits a limit as the number of cells is increased. 4UU 300 200 100 0 0 2 1 3 Vds [V] Fig. 5.1 DC characteristicsof the 90-nm device (1x48x16 lim) with indicatedload-lines. The model must focus on the device regions that the load-line occupies. The DC data shows self heating effects for high Vds, Id values. 5.1. Model Structure The basic model is shown in Fig. 5.2. The model contains 7 parameters which are described next. The voltage-controlled current generator requires three parameters. It consists of the threshold voltage (VT), below which the device is non-conducting. The exponent term (exp) is needed because the transconductance is not constant with bias (we will discuss this in section 5.2.). The current term is proportional to the factor k. The maximum current will be limited by the device's on-resistance (Ron), and the breakdown voltage (BV) is implemented as a diode that turns on sharply at BV. The model also includes the input and output impedance networks. They consist of pad, interconnect and intrinsic gate-to-source and drain-to-source parasitics, as well as the output conductance of the intrinsic device. In a load-pull setup, we can assume that the reactance of the device is resonated out by the impedance matching network, leaving only the resistive components. Therefore, for the purposes of RF power estimates, the aggregate of the gate and , - , - Dain R.O NII Fig. 5.2 Full device model (top) and simplified equivalent model (bottom). The model is a large signal model. The R-C networks on the drain and gate are simplified into a single frequency dependent resistor. The assumption is that the reactive components of the device are resonated out by the source and load impedances. The other model elements include the current generator(k, VT, exp), the breakdown voltage (BV) and the on-resistance (Ron). drain parasitics can be modeled as a frequency dependent resistor, RGS = Real{ Zs(f) } and Ro = Real {ZDs(f,Bias) }. 5.2. Parameter Extraction Model parameters are extracted with the help of DC and S-parameter measurements. DC measurements of the output and subthreshold characteristics allow for extraction of VT, BV, Ron as well as k and exp. Typical values for the 65 nm and 90 nm standard devices are shown in Table 5.1. The parameters for k, exp and VT are extracted by matching the subthreshold curve at Vdd=lV, shown in Fig. 5.3. The parameters are treated as matching parameters, with the goal of providing an accurate match of the subthreshold current around VT. The on-resistance (R.o) is extracted from the measurement with V,s=l V, Vds=0.01 V, where Ron = Vds/Id. The RGS and Ro values are extracted from the output impedance of the device at the frequency of interest. For this, in Fig. 5.4 the output resistance extracted from small signal measurements, 1/Real(Y22), is plotted as a function of VGS for VDos=VDD. We observe that this parameter is strongly bias point dependent. For low VGS, the device is turned off and Ro is large. For values of VGs below threshold, Ro rapidly drops and then levels off. Our simple model requires a fixed value for Ro. The load line sweeps through different regions of device operations (Fig. 5.1). This makes the choice of Ro challenging. We have found that a good choice for Ro is around the class AB bias current (indicated in Fig. 5.4). Using a non-uniform value of Ro is difficult to implement in the simulation environment, because of discontinuity concerns [26]. It would also introduce additional complexity to the model. The extraction of Rgs is easier because there is no strong bias point dependence (Fig. 5.5) unlike the case for Ro. Therefore, Rgs simply is the input resistance to the gate, 1/Real(Yi ), measured at the appropriate frequency and Vds=lV, Vgd=VT. Parameter 65 nm Device VT (V) K (S) Exp 0.27 0.39 90 nm Device 0.40 0.55 2.32 1.60 Ron (Q) BV (V) 2.3 3.5 2.3 Rgs at 8 GHz (Q) Ro at 8 GHz (2) 246 120 4 215 110 Table 5.1 Model parametersextractedfrom DC measurements,for the 65 nm and 90 nm technologies. 1 0.1 0.01 0.001 0 0.5 1 Vgs [V] Fig. 5.3 Example fit of the subthresholdcurve at Vds= 1V, for the 90 nm device. The extracted values for VT, k and exp are shown in Table 5.1. 10000 1000 100 10 1 0 0.2 0.4 0.6 0.8 1 Fig. 5.4 Measured output resistance (Real(Y22)) as a function of Vgs, for two frequencies at Vds=1V. The data shows that Ro, is bias dependent. The model assumes a bias-independentRdos. Choosing a constant value around threshold (indicated)provides the best compromise. - IUUU values used in model r 1000 - 2 GHz 8GHz 100- 10 - I 0 I I I 0.2 0.4 0.6 0.8 Vs IV] Fig. 5.5 Measured input resistance (Real(YDl)) as a function of Vgs, for two frequencies at Vds= V. The data shows that Rgs is largely bias independent.Model parameterswere chosen at Vgs= VT. 5.3. Model Results Model simulations were performed at ADS. Because the model assumed that reactive components can be resonated out, the lack of capacitances or inductors allowed the actual ADS simulation to be done at an arbitrary frequency. The model Rgs and Ro were set according to the power device frequency that was simulated. The simulation optimized for both peak PAE and output power, using Vgs, Rsource and Rload as parameters. Fig. 5.6 compares the ADS simulations and measurements of the peak PAE and output power as a function of frequency. The general trends of the measurements are present in the simulations, such as the frequency independence of the output power (until the gain approaches unity at very high frequencies) and the drop of peak PAE with frequency. Also, both the 65 nm and 90 nm devices show very similar model performance. We can now apply the model to the question of device scaling. Fig. 5.7 shows the model performance for different values of parasitic resistance for the 65 nm scaling experiment discussed in Chapter 4. As the simple model of Chapter 4 suggested, the peak PAE is dropping with the number of parallel devices, while the output power rises and eventually levels off and decreases. The model is different from the analytical one introduced in Chapter 4. The circuit model presented here relies on device predictions using harmonic balance simulations. It is able to properly take the full impact of the parasitic backend resistances into account, such as source degeneration for example. The model only required DC and S-parameter measurements, which are readily available. It does not require RF power measurements. The analytical model in 100 80 15 E a1. 60 10 0 .O. 40 5 0 20 0 0 20 15 10 5 Frequency [GHz] Fig. 5.6 Model performance comparison with measured power performance, as a function offrequency. The model uses the parametersof Table 5.1. The model for each technology roughly matches the behavior of the data. The output power is frequency independent up to very high frequencies, while the peak PAE decreases. 250 45 200 o30 150 &a I. Cz 7 E 100 w n LL· 50 0 0 2 4 6 8 Number of Cells (Nc) Fig. 5.7 Model performance compared to the 65 nm device scaling measurements. Depending on the parasiticresistancefor the overall device, the drop in peak PAE and the maximum output power can vary significantly. The model assumes a constant backend resistance that does not scale with the number of cells. Chapter 4, on the other hand, is based entirely on adjusting the output power (Pojt) and PAE equations, using only the RF power measurement results. The simple circuit model presented here also highlights the sensitivity to the parasitic resistance, as seen in Fig. 5.7. If the backend contains a fixed 0.25 Ohm resistance associated with the device pads (it is independent of Nc) in the source, drain and gate, the model predicts peak power levels above 100 mW. However, if this resistance increases to 1 Ohm, the power will be limited to less than 30 mW. This model allows us to estimate the maximum output power that can be achieved in a given technology. It is clear that the limiting mechanisms are the parasitic resistances, which dissipate power and result in an I-R drop. In the best case scenario, the lowest resistance is achieved with a metal plane above the device directly connecting to the bond pad. Assuming a square device area, this would require a minimum of two to three metal squares. The sheet resistance of a thick metal layer is approximately 25 mQ/o. Therefore, we can expect about 75 mQ of parasitic Rs and Rd. Another contributing resistance would be the bond wire resistance, which is estimated at around 103 mQ [27] for a 2 mm long wire of 1.0 mil diameter. It is therefore conceivable that the lowest parasitic resistance is about Rs=Rd=Rg=180 m92. Extrapolating Fig. Fig. 5.7, this would place the maximum power for the 65 nm device to about 100-150 mW. At the maximum power level, the peak PAE will have dropped significantly from its single-cell value (the simple model suggested a drop to ½2 its original value in Eq. 4.8). In Figs. 4.5 and 4.9 we saw the peak PAE decreasing with the number of cells. Design requirements may dictate a minimum acceptable value for peak PAE. Therefore, this also places a limit on the number of parallel cells, consequently also limiting the total output power to less than its maximum possible. This hold especially true for higher frequencies, where the single-cell peak PAE is already quite low (see Fig. 4.13). We would like to find the maximum output power at a given frequency, subject. to a required peak PAE level. Clearly, as the demands on peak PAE rise, the maximum output power at that PAE level will be reduced. The next chapter will discuss how we can develop a picture that is able to capture all of these tradeoffs. Chapter 6. Discussion Chapters 3 and 4 have shown measurements that establish the trends of the output power and peak PAE with frequency and multi-cell scaling. The models developed in chapter 4 and 5 help us to formulate the trends seen in the measurements. This chapter will draw on this to derive empirical expressions for the output power and peak PAE, as a function of device size and frequency of operation. This will allow us to derive a compact picture that helps us to understand the impact of CMOS scaling on the power performance. From Fig. 3.3 we saw that the peak PAE is roughly linearly decreasing with frequency (f), extrapolating to zero at a maximum frequency (f,,t), and roughly its class-B theoretical value (id4) at DC. We can capture this as: PAE =- 1 Eq. 6.1 This is the efficiency for a single cell device. If we start to place multiple devices (Nc) in parallel, the peak PAE will drop even further due to the parasitic resistance as discussed in chapter 4. Depending on the parasitic resistance and its scaling behavior, there is a maximum number of cells in parallel beyond which the output power is no longer increasing (Nsat). Fig. 4.5 and the simple model described in chapter 4 suggest that the peak PAE will drop linearly with the number of cells. At Nc = Nsat, the peak PAE has dropped by half, so that for Nc = 2 Nsat it is reduced to zero. Therefore, we can modify Eq. 6.1 to take this into account: PAE = - - t 1 Eq. 6.2 sat Similar considerations can be done for the output power. Fig. 4.13 demonstrated that the output power is independent of frequency. Therefore, it will only be impacted by the number of parallel cells, and the parasitic resistances. A single cell device produces PNc=1 of power. At Nc = Nsat, the power has saturated to the maximum possible (Psat), which can be found either from the measurements in chapter 4, or the model results from chapter 4 and 5. A first order expression of the simple model in chapter 4 was found as: Nsa, 2 V• clooRN (Eq. 4.7) and Psat, = RN PAEo 4cR, (Eq. 4.8) We can use these equations, or choose values based on the measurement observations if a more accurate fit without an analytical solution is needed. Based on the simple model in chapter 4, and the more detailed model discussed in chapter 5, we will assume that the power increases from PceIn to Psat in a quadratic way. Therefore, we can capture this as S ce Na a1 -N Eq. 6.3 This expression captures the peak in output power, and for Nc>Nsat, the power starts to decrease, as seen in Figs. 4.5 and 4.9. We can now use the equations for peak PAE and output power (Eq. 6.1 and Eq. 6.3) to graph constant PAE contours in the power-frequency space. We can combine the two equations to give a single expression for output power, using frequency and desired PAE as its variables: Pout Eq. 6.4 PNc= + (P,, - P,, X)(1 - a2) with - PAE 1 a 11-1/Nsat - This is shown in Fig. 6.1. For a fixed level of PAE, as the frequency increases, the maximum possible output power will decrease, as Fig. 6.1 shows. From the figure, we see that as the frequency is increased, the peak PAE gradually drops and will eventually reach zero at feut. The output power shows an upper limit, set by Psat. It is important to note that when approaching this limit, the PAE rapidly drops. In fact, Eq. 6.2 and 6.3 illustrate this: the power reaches its maximum at Nc=Nsat. At low frequencies, the PAE from Eq. 6.2 then simplifies at its peak power level to: PAE(Pa )= - Na= 4( 39.2% Eq. 6.5 2Nsat4 Therefore, unlike the frequency limit at which PAE gradually dropped to zero, the maximum power limit at low frequencies shows a peak PAE of 39.2%. This can be seen in Fig. 6.1, as the sharp drop of PAE as the output power reaches Psat. ~nnST~nT e'-U ~~r_ C~)rllOLlr~i 180 160 ,140 E 120 3 100 o LL80 a 60 40 20 1 10 Frequency [GHz] 100 Fig 6.1. Constant PAE contours using the model, illustrating the output power that can be achieved at a given frequency. The higher the desired PAE, the lower the power andfrequency. The power has an upper limit set by the backend parasitics,while the frequency has an upper limit related to the device gain andf,,. The values in the graph are based on the 90 nm technology, which we 2 expect to be able of delivering 140 mW of power (Psat) with a maximum power at Nc,sat=1 , and a maximum operatingfrequency of 21 GHz (fsat). 6.1. Frequency and Power Limits of 65 nm, 90 nm and 250 nm technologies With the insights developed in the previous section, we can now complete the picture of the CMOS application space from the introduction (Fig. 1.9). Both from chapter 4 and the model results shown in Fig 6.1, we can get an estimate on the maximum power level that can be achieved by combining multiple device cells. For the 90 nm node, this is estimated to be about 140 mW at PAE = 50% for low frequencies, based on the measurements shown in Fig. 4.5. The 65 nm node saw a peak power level of about 50 mW (chapter 4). From the literature [28,29] and Fig. 1.3, we estimate the 0.25 jim technology to give about 1W. The frequency limit for a single cell device in the three technologies is estimated to be roughly identical. We draw this conclusion from the measured fmax for the standard devices, seen in Fig. 3.7 and Fig. 6.2. We can now estimate the power capability. Fig. 6.3 shows the 50 % PAE contours in the frequency/power space, for 65 nm, 90 nm and 250 nm technologies. As the technology is scaled down, the maximum output power decreases due to the lower Vdd (consistent with Fig. 1.3). The power capabilities of 65 nm still include a good portion of the application space, but leave out the higher power applications. The backend layout issues seen in Chapter 5 cause the maximum power level in the 65 nm node to be lower then that of the 90 nm node. WG,TOT = 768 pm 50 Id = 26 mA/mm Optimized Layout Vdd = Vnominal 40 '1(D ,6 30 S20 0L10 65 nm 90 nm Gate Length 250 nm I Fig 6.2. Impact of device layout on the device f,. The standard single cell device is shown in the solid bars. Using an optimized device layout by breaking the device up into multiple smallercells, fa can be improved significantly. This becomes more critical as the technology is scaled down. I I h 1 0.1 0.01 1 10 100 Frequency [GHz] Fig 6.3. Estimated contour of 50% peak PAE, for three different technologies, illustrating the power vs. frequency behavior. The estimates are based on extrapolationof measureddata. In chapter 3 we saw the impact of the device layout on fmax. Fig 6.2 also includes measured fmax data for the best layout we fabricated (using many smaller devices in parallel as opposed to s single cell). Here, we can see that if we re-optimize the single cell standard device, the 65 nm technology shows the largest potential for pushing into higher frequencies. For the 0.25 Am device, we expect the layout decisions to be less important, because the unit cell is relatively small with respect to the intrinsic device dimensions (the gate aspect ratio, WG,F /Lg in only 80:1 for the 0.25 Am device, as opposed to about 230:1 for the 90 nm device). If using an optimized layout, then the maximum frequency limit will improve as we scale the technology. This leads us to revise Fig. 6.3 into the results shown in Fig. 6.4. We expect the output power limit to decrease as technology is scaled. This is primarily due to the Vdd scaling. Because the nominal voltage between the 65 nm and 90 nm devices has changed only marginally (both quote Vdd=l V but the 65 nm device has a somewhat lower breakdown voltage), their output power performance should be very similar. On the other hand, the 250 nm device has a much higher Vdd of 2.5 V, and is therefore able to produce higher output power levels. This picture is consistent with Fig 1.3. The frequency limit can be pushed out with scaling technology. However, in order to realize this improvement, careful layout design considerations have to be made, because the reduction of Vdd results in an increase of bias current and vulnerability to I-R losses in the backend, as we saw in chapter 4 and 5. Based on our measurements and the model estimates, we were able to identify the trade-offs that exist when scaling CMOS devices for RF power. The reduction of Vdd will directly translate into a lower output power. On the other hand, device scaling increases the speed of the intrinsic device, which can translate into higher frequency performance. But in order to capture this 1 o a. u- 0.1 U.01 I 1 I I 1 I I III I 10 I I Iiir~l 100 Frequency [GHz] Fig 6.4. Estimated 50% peak PAE contours,from Fig. 6.2, redrawn taking the device layout optimization into account. Layout optimization can help to push out the frequency limit for the PA device. The 65 nm technology profitsfrom optimized layout the most, as seen in Fig. 6.3. improvement in the frequency response, a careful layout of the device is essential. Layout is also critical in reducing parasitic resistances that can substantially limit the possible output power. We demonstrated that the device layout decisions are becoming increasingly important as CMOS is scaled down, substantially reducing the output power if not properly optimized. If the layout is properly taken into consideration, the loss of output power due to multi-cell scaling can be lowered, and scaling will push out the operating frequency. The loss of power density due to a lower Vdd, however, is fundamental, and therefore scaled CMOS will always result in a reduction of the maximum possible output power. Chapter 7. Conclusions and Recommendations We have studied the RF power suitability of CMOS technology from 250 nm to 65 nm. Both the 65 nm and 90 nm technologies have a nominal voltage of 1 V. At a bias of 1 V, the 90 nm nominal devices offer the best output power and peak PAE, but 65 nm devices closely match it. The 90 nm technology achieved a peak PAE of 59% at 8 GHz, with a power density of 28 mW/mm. The 65 nm technology achieved a peak PAE of 53% at a similar power density of 29 mW/mm at 8 GHz. If one allows for optimum voltage selection, the 250 nm thick-oxide devices offer the best output power and efficiency at the highest voltages, with 59% peak PAE at 124 mW/mm output power density at 8 GHz and Vdd = 2.5 V. We have shown that the device layout has a large impact in the power performance. A single cell 65 nm device exhibited a peak PAE of 62% at 4 GHz. Breaking it up into many smaller cells, but of the same total width, increased the peak PAE to 66% while maintaining the same output power. We also demonstrated that in order to increase the overall output power, scaling the device width is very effective. This is best done by wiring together many smaller unit cells. However, as the device size increases, parasitic resistances start playing an important role and eventually limits the maximum output power that can be achieved through width scaling. We also showed that as the device width increases, the peak PAE decreases due to the loss in the parasitic resistances. This results in a tradeoff between output power and peak PAE, as we change the device width. Another important tradeoff we have shown is the frequency dependence of the power performance. The peak PAE drops roughly linearly with frequency, while the output power is independent of frequency over a broad frequency range. All three technologies are capable of amplifying signals up to about 20 GHz, although the PAE drops significantly as the frequency approaches this boundary. Through measurements and model predictions, we were able to analyze how gate length scaling impacts the power performance. We found that the reduction of Vdd associated with scaling causes the output power density, and therefore the output power, to drop. Lowering Vdd also makes the device more susceptible to parasitic resistances, because for a constant amount of power, higher currents are needed. Device scaling increases the speed of the intrinsic device, which translates into a higher fax and higher PAE at a given frequency. But in order to capture this improvement in the frequency response, a careful layout of the device is essential. Layout is also critical in reducing parasitic resistances that can substantially limit the maximum possible output power. We demonstrated that the device layout decisions are becoming increasingly important as CMOS is scaled down. If the layout is properly taken into consideration, scaling can be instrumental in enhancing the operating frequency of CMOS. Designers using a deeply scaled CMOS technology have to decide which device to choose for the power amplifier. The first question to consider is whether thicker oxide I/O0 devices or higher voltages are available in the design. This is mostly a question of cost, likely to be determined by other system components. If thick I/O devices are available, it makes sense to choose them for lower frequency applications, such as 2-5 GHz, because of their higher power capability. If the frequency of operation is approaching the limits of the thick-oxide devices (1020 GHz), the design can benefit by using the nominal thin-oxide devices, because they offer better frequency performance. If both the frequency and power demands on the application are very relaxed (e.g. in a low power bluetooth design), the choice of the device type should be guided by non-technology factors, such as existence of previous designs, cost or convenience. Future research can focus on many issues discussed in this thesis. Device layout decisions are very important. However, a most important contribution to understanding the capability of CMOS would be to populate the frequency-power tradeoff estimates in Fig. 6.1 with actual device measurements. If this is done for different technologies, designers could understand and compare technologies and their limitations, by using a set of very simple but powerful figures. The results of each technology could be combined into a graph similar to Fig. 6.4. In Chapter 4 we saw that the output power of the 65 nm device was limited by layout issues in the width scaling. With proper layout, we expect the 65 nm device's output power performance to be similar to that of the 90 nm device. This is indicated in Fig. 7.1. Populating Fig. 7.1 with actual measurements, a single graph would be able to show the benefits and tradeoffs of different CMOS technologies. Ultimately, one could also include alternative technologies to CMOS, such as SiGe HBTs or III-V semiconductors to provide a comparison on an even broader choice of technology. __ 10 i ,· a3U- n- 0.1 0.01 1 10 100 Frequency [GHz] Fig 7.1. Expected 50% peak PAE contours,from Fig. 6.4. This figure assumes an optimized backend layout for the 65 nm devices, resulting in an improved width scaling. We expect the maximum power of the 65 nm devices to come close to that of the 90 nm devices. Appendix A Measurement Theory This section describes the measurement configuration and calibration procedures. In a single tone power measurement, shown in a simplified form below (Fig. A.1), we are typically interested in the following measurement parameters: is not The available RF input power to the device under test (DUT). If the device perfectly matched, some of this power will be reflected back to the source, and the actual Pin power entering the device will be reduced. Pout Vgs The RF output power that leaves the DUT The gate-source bias (or Vbe for BJTs). This is the voltage applied on the source side of the device, also is commonly referred to as Viow Ig The current entering the gate. Also referred to as I'ow Vds side of the The drain-source bias (or Vce for BJTs). This is the voltage applied on the load device, also is commonly referred to as Vhigh Id The current entering the drain. Also referred to as Ihigh Drain Bias Network Gate Bias Network V DUT RF Input Power DC Supply: Gate RLoad DC Supply: Drain Load Fig.A.1: Basic setup of the power measurementsystem. The device (DUT) is biased with a bias network ("Bias-Tee") and RF power is inserted into the gate. The load is connected on the drain. From the above parameters, we can compute several figures (all numbers in SI units, not dB or dBm): Gain The power gain of the RF output power, relative to the available RF input power: Pt (unitless) Gain = "t Pin The drain efficiency, a measure of how much DC power is converted into RF output 1r power: (in %) Pout S out - gVgs +IdVds Doc PAE The power added efficiency, a measure of how much DC power is converted into RF output power, taking into account the RF power that enters the device: PAE= o-u in Poc out, n gV + Id Vds Pou - Gain) IgV, + IdVds If the power levels are expressed in term of dBm, the above become: (in dB) Gain = Pout - Pi = 10P,,/ 0-3 gg (in %) d+ IVg, ++IdV PAE - 10 *,,1o-3 _-104,/10-3 IV•, +IdVd (in %) (in %) At RF frequencies, however, the setup in Fig. A.1 is incomplete. We need to take into account the fact that the source and load impedances have a significant impact on the device performance. Therefore, adjustable tuners are inserted right before the gate (source) and drain (load) of the device. In addition, a directional coupler is placed after the RF source, so to get a more accurate reading of the source power. The complete schematics of the source and load side of the measurement system are shown in Figs. 2 and 3, including RF cables: rigner mr-loss results in worsening of the system's maximum achievable VSWR. Cable I Connector SEquipment Fig. A.2: The system setup on the source side. The RF inputpower is read by using a directional coupler. The RF power is then combined with the DC bias. The source tunerpresents a variable impedance to the DUT. Cable Connector worsening of the system's maximum achievable VSWR. Equipment Fig. A.3: The system setup on the load side. The load tunerpresents a variableimpedance to the DUT, and the RFpower is measuredat the RF end of the Bias-Tee. Measurement De-Embedding Out of the 6 measurement parameters mentioned in the previous paragraphs (Pin, Pout, Vds, Vgs, Id, Ig) we can only directly observe Idand Ig.The methods and challenges in observing the other parameters are described below: The voltages Vds and V, are measured at the DC bias source, and therefore any I-R drops between the DC source and the DUT will reduce the actual voltage at the DUT plane. Using a "force-sense" setup (one probe delivering current, and a second probe measuring voltage) is not possible, the voltage sensing probe would deteriorate the RF matching to the device. However, for small currents, the I-R drops in the measurement cables are insignificant. This is, of course, particularly true for the gate - where Igis negligible in this context. On the drain side however, for very large devices that require high current at low Vds, this might become a concern. The measurement system cannot offer a solution to this problem, other than keeping the cable length between the Bias-Tee and the DUT plane at a minimum. In order to find the input and output power to the device (Pin, Pont), we need to realize that the power measured (Figs. A.2 and A.3) differs from the true Pin and Pot due to the RF elements inserted between the DUT plane and the power meter. However, all of the RF blocks between the DUT plane and the power meter are linear 2-port elements. We can use S-Parameter analysis to de-embed their effects, and obtain the true value of the power levels at the DUT plane. The computational details are described in the next sections. De-Embedding and S-Parameter Analysis The load side of the measurement system is shown in Fig. A.4 below. The power leaving the DUT enters into a series of cascaded S-parameter blocks. These are: the fixture (which includes the probes and the cable connecting them to the tuner), the tuner itself, and the Bias-Tee (and any cables connected to it). The power meter terminates this series, and is characterized by its reflection coefficient (Fpower-meter = Si 1). Tuner Fixture Bias T and Power Meter (Probe and Cable) Cables Sl Fig. A.4: The load-side of the setup, as a series of cascaded S-parameter blocks. The port numbers are indicatedabove the 2-port blocks. To simplify this picture, we will first combine the 2-port S-Parameter blocks into a single equivalent block. Cascading two S-parameter blocks Two S-Parameter blocks connected in series can be combined into a single block by using chain scattering parameters (T-Parameters). The conversion between S- and T-parameters for 2-port blocks is shown below [1]: Ts_(S )= [ S 12 S 21 - S 1 S 22 S22 ST-4S T T i2) Ti1 S 1 T22 - Tl2 -T21 •T2 lx S21 1 T22 Fig A.5: Using T-parameters, the two cascaded S-parameter blocks (SA and SB) can be combined into a single S-parameterblock (SA+B). We can make use of the fact that the T-parameters of two blocks connected in series satisfy: TA+B = TA .TB Cascading two S-parameter blocks therefore is done by converting them into T-parameters and performing a matrix multiplication of the two. The result is then converted back to S-parameters: Cascade(SA, SB) = SrTs (TS-T (SA) -Ts T (SB)) This result now allows us to simplify the setup that was shown in Fig. A.4, the resulting setup is illustrated in Fig. A.6: S Equiv = Cascade(SFture, Cascade(STu,,er, S B•,,-)) = STT-S (Ts--T (SFixture ) - Ts.T (STuner ) - Ts•T (SBiasT )) Equivalent Power Meter S-Parameters S11 Fig. A.6: Cascading all of the 2-port S-Parameter blocks of Fig. A.4, we can simplify the measurement system into the DUT connected to the power meter through a single S-Parameter block. The port numbers are indicatedabove the 2-port block. Input impedance of a terminated S-parameter block Before we can find the power that leaves the DUT, we need to compute the impedance that is presented to the DUT in Fig. A.6 (by the S-parameter block terminated by the power meter). The equivalent reflection coefficient of the setup shown in Fig. A.7 is found to be [2]: Equiv = + S12 S S 1- Sz, _s2, Load S12] S22] S22•Load ** rLoad -- -Equiv Fig. A. 7: Finding the impedancepresented by an S-parameterblock terminatedon the 2 nd port. Finding the amount of power delivered into the load For the network shown in Fig. A.8, we can compute the power delivered to the load, given the power that leaves the DUT [3]. The ratio of the two will give us the gain (or attenuation) between the two power levels: Equivalent Power Meter S-Parameters Sl Fig A.8: Block-setup for computing the power delivered to the load. m 2 1 _ IS2Ll IS21 PLoad avail - , )x1- S 2 1o) with S11 , S12, S21, S22 are the S-Parameters from SEquiv, and PLoad is the power delivered to the load, and Pavail is the power that is available from the DUT. FLoad is the reflection coefficient (S 1) of the power meter. FIN was defined in the previous paragraph, as FIN = = Fuiv (S, Load ) S 12 21 Load 1- S22* rLoad The gain in dB between the measured power (PLoad) and the power leaving the DUT (PAvail) is: Load Gain = GAT, (S , FIad ) = 10 x log Pa I availi IS21 2 X(1- IrL' I) = 10 x log 1- S 1-2 21 12 I- S22 Load Load x - S221Load )1 Therefore, at the device plane, the power level is expressed as Pout = POut,DUT POut,Measured - -Gain= POut,Measured - GAT (SEquiv' Load ) Where POut,Measured is the power meter measurement. The power leaving the DUT (Pout) is larger than what is read off the power meter (POut,Measured) because of the loss and mismatch of the elements between the DUT and power meter. The Gain in the above equation is a negative dB number. Calculating the Input Power at the Device Plane The input power de-embedding follows a similar concept. Fig A.9 shows the S-parameter schematic of the source side of the measurement system. The power available from the source is measured through the directional coupler (see Fig. A.2). Given the source power, we find the power entering the DUT similar to the approach used for the load side. We first combine the 2port S-parameter blocks into a single equivalent block (Fig. A.10). 1 2 Power Source Bias T and 2 1 Tuner 2 1 Fixture (Probe and Cable) S,, Cables S-parameter blocks. The port cascaded of a series as the setup, of source-side Fig. A.9: The numbers are indicated above the 2-port blocks. Power Source Equivalent Sl S-Parameters Fig. A.1O: Combining the S-parameterblocks of Fig. A.9, to simplify the source side setup. This setup is very similarto the one presented in Fig. A.8. The gain in dB between the source power (Psource) and the power available to the DUT (PAV,DUT) is, similar to equations [XYZ]: Gain = Garr (SEquv"'Fsrc)= 0lx log PSource PAv,DUT SrcI2) S2 12XX(1 =10xlog 1-S+ S12 S21 1- S22 - src Src x 1- S22 Src I The difference in this analysis with that of the output power is that we are working in the opposite direction: given the power at the termination (Source: power source S 11), we need to find the power into the network (PAV,DUT). This is inverse of the output power problem, where we needed to solve for the power at the termination (Load: power meter S11) given the power into the network (Pout,DUT). Therefore, the power available to the device is lower than the power available from the source, due to mismatch and loss. The equation for the available input power (PIN = PAv,DUT) to the device then becomes: P, = PAv,DIUT = PSource,Measured + Gain = Psource,Measured + GA E(SEquiv ource Where Psource,Measured is the available power from the source, measured through the coupler in Fig. A.2 (or, in alternative system setups, known through calibration). Finally, we need to relate the source power (PSource,Measured) to the actual reading of the input power meter in Fig. A.2 (PIn,Measured). The two are different because the power meter is reading the actual source power through the connection of cables and the directional coupler. Therefore, the power arriving at the power meter will be attenuated, from the original source power level. We will call this attenuation the coupling between the power source (PSource,Measured) and the power meter (PIn,Nleasured): Coupling = PSource,Measured - Pn,Measured Thus, the input power equation can be rewritten as: PI,, = PIn,Measured +Gain -Coupling = P=n,Meaured +GA,(S qsource )- Coupling Determining the Source Coupling The coupling factor is found through calibration. The probes are connected to a through line, with known (measured) S-parameters. We can then compute the input and output power at the device plane shown (Fig. A.11), which will refer to exactly the same quantity. A simplified schematic is shown in Fig. A.12. Both power levels must be equal, and we can compute the coupling factor. Probe Probe Si Power. 5" S1,,2SIS21 S2] Source Power Bias T and S11 Cables j F. S [S S,,S ,1 S22 S22 S12 SS21 Fixture Source Tuner (Probe andCable)' ,, S12 22 S2' S22 Through Line S , S, LS Fixture (Probe and Cable) S , S,1 Load Tuner 5, S,2 SSi POW Bias T and Power Meter Cables S11 Fig. A.11: S-parameterschematic forfinding the couplingfactor. The power crossing the dotted line from the source to the load side is identical to Pin as well as Pout. We can thus set up equationsto solve for the coupling. Probe Probe I S1i Power s-s,,s,,s,,sl S Source S11 [--* 21 2 1 [S1 S2 12 I 2 S Power Meter Power Source Equivalent Equivalent Power Meter S,, Source Load S11 Fig. A.12: Simplified S-parameterblock setup of Fig. A.11, by cascadingthe 2-port networks. First, the equivalent S-parameters for the source and load side are found using the S and T parameter transforms: SEquivSource = Cascade(SFxuresource, Cascade(STunerurce , Saourc )) = S'TIS (TSarT (S xuriSource ).-TS-+ (STunerSource ) Tsar(SBiasTSource)) SEquivad = Cascade(SThroughne, Cascade(S = ST-S (TS+T (SThroughine ) Cascade(STunerLoad, SBiasM)ad )Fturead, TST (S itureLoad) TS-T (STunerLoad ) TS-+T (SBiasTLoad)) Finding the power levels at the plane of reference, from both sides, gives: PRight = POut,Measured - GATr (SEquivLoad 9SI ,PowerMeter) PLeft = P,easured + GAT (SEquivSource S,PowerSour )- Coupling Where Pin,Measured and POut,Measured are the power readings from the power meters (Fig. A.2 and 3), SEquivLoad and SEquivSource are the combined S-parameter blocks from the equations above, and Sll,PowerMeter and S11,PowerSource are the reflection coefficients (S 11 measurements) of the output power meter and the power source. The coupling factor is found to be: Coupling = PIn,Measured - POut,Measured + GATT (SEquivSource S lPowerSource)+ GATT (SEquivLoad 'S 1.PowerMeter) We have now found equations relating the readings of the input and output power meters to the actual power available to the device (Pin) and the available power leaving the device (Pout). These values are to be used in the computation of Gain, PAE and rid shown in be beginning of this chapter. Measuring the Reflected Power In a general load-pull system, the reflected power from the gate of the device can also be measured. For this, an additional power sensor needs to be inserted into the source side through a coupler. This sensor will couple and measure the power reflected from the gate back into the source. The mathematics for determining the reflected power will be identical to that of the output power, as the setup will be identical to that shown in Figs. A.3 and A.8, with the insertion of a directional coupler. To avoid having to create a 3-port model for the coupler, the power meter reference plane (equivalent S11 and loss terms used in the setup shown in Fig. A.8) is best established directly after the tuner. This typically coincides with the power source S11 (unless the bias Tee and cables are treated as separate S-parameter blocks). References Ed., Wiley, pp. 206-212 [1] D.M. Pozar, "Microwave Engineering", n 2 d [2] D.M. Pozar, "Microwave Engineering", 2 nd Ed., Wiley, p. 216 [3] D.M. Pozar, "Microwave Engineering", 2 nd Ed., Wiley, pp. 606-609 100 Appendix B Measurement Equipment This appendix lists the equipment used for the measurements and their manufacturers, as well as a detailed system configuration: Equipment List Note: Not all equipment is used at the same time in the setup Eauioment Load Pull System 1.8-18 GHz Tuners Manufacturer PartNumber(s) Tuner Controller Measurement Software Additional Measurement Software MT982A02 Maury Microwave MT986B02 Maury Microwave Maury Microwave Customized Software (written in VB.net) Probe Station 8" Manual Probe Station Noise Isolation Table (Airtable) RF probes (GSG) Cascade Microtech Vibraplane GGB Industries (PicoProbe) Summit 11102B Kinetic Systems 40M-GSG-126-PLL 2-18 GHz Amplifier 2-4 GHz Amplifier 4-8 GHz Amplifier 8-18 GHz Step Attenuator HP Hughes Hughes Hughes HP 2-18 GHz, High power Agilent 8672B 1177H01 1277H02 1177H15 11713A, 11716C, 8494H,8496H, 8120-2703 E8257C Opts 1El & 1EA Power Source Cables Network Analyzer Calibration Cables Probe-to-Tuner cables All other cables DC (coax) Triax-to-coax DC adapters Gore GDOAKZ3G0520 Gore G2KOD01008.0 Semflex Y1N3HPT19036 0419, M119BFSM300360 414, M119BFSM30018 0414, M119BFSM30072 0415, M119BFSM30024 0414, N119BFSN10018 0414, M219BFSM10018 0447 Trompeter Off-the-shelf ADBJ20-E2-PL75 0420 Trompeter 101 Passives Directional Coupler RF Isolator 2-4 GHz RF Isolator 4-8 GHz RF Isolator 8-18 GHz Bias-Tees Narda Microwave Narda Microwave Narda Microwave Narda Microwave Agilent 5292 4913 4914 IGS-8018 11612A Maury Microwave Maury Microwave Maury Microwave Maury Microwave Maury Microwave Maury Microwave Maury Microwave 8022B 1 8022A1 2625B 7927D 2606D 2606C 8023D1 HP HP Agilent Agilent Agilent 6628A 4145B E4419B E9300A E9300H RF Adaptors 3.5mm(m)-ACP7 3.5mm(f)-ACP7 SMA(m)-ACP7 2.4mm(m)-3.5mm(f) N(m)-ACP7 N(f)-ACP7 N(m)-3.5mm(m) Other Major Equipment DC Bias Source (drain) DC Bias Source (gate) Power Meter (dual channel) Power Sensor (20 dBm, Input) Power Sensor (30 dBm, Output) Source Configuration A: Low Power RF Source I 8" RF Source 8672B i"~-~r~ WT PA 127"I-12 ~?? I I Power Meter E4419B, E9300A S Isolator IGS-8018 8022B1 Sr RFProbe Tuner 40M-GSG-125-PLL ourT982A2 8022A1 G2KOD01008.0 Fig. B.1: Source-side setup for 8-18 GHz. 102 Source Configuration B: High Power RF Source I I Power Meter E41B E30 == -m 24" 8022A1 -I I RF Sourc 8022A1 E8257C 18" M219BFSM10018 RF Probe 8022A1 140M-GSG-125-PLL ADBJ20-PL75 Fig. B.2: Source-side setup for 1.8-18 GHz. Power Meter E4419B, E9300H I Y1N3HPT19036 I 8022B1 RF Probe LoadTuner 40M-GSG-125-PLL MT982A02 G2KOD01008.0 8022A1 Fig. B.3: Load-side setup. 103 Input PowerMeter E4419B, E9300A Y1N3HPT19036 Reflection Power Meter E4419B, E9300A s3" Y1N3HPT19036 Dual Directional Coupler 8022A 24" _ _ _-- _"_ OUKK0024.0 8022A1 s18" M219BFSM10018 Plane for Power Meter S,1 and Loss I -j _ RF Probe 40M-GSG-125-PLL Source Tuner MT982A02 Bias Tee 11612A G2KOD01008.0 8022A1 _ _ Bias _Gate ADBJ20-PL75 Fig. B.4: Possible modification of the source-side setup for 1.8-18 GHz to include a reflected power measurement. The calibration plane for the reflection power meter is indicated. 104 Appendix C Device Library 65 nm Devices This section gives an overview of the devices fabricated in 65 nm technology for this work. Sizing considerations as well as groups of possible sets of experiments are listed. Devices are have an identifier number (id) to help locate them on the wafer. A map of the device locations is shown at the end of the section. Determining Device Sizes The center device is sized to achieve a certain power range. This range is computed under the following assumptions: 1. desired output power = 100 mW (20 dBm) 2. 3. at a PAE of 50%, and Vdd = 1 V, this requires a DC current of 200 mA this current is estimated to be to 1/2 (class A) to 1/3 of In. Thus an Ion of 400 mA is needed with a current density of Io = 475 gA/jm, this corresponds to a width of 842 gim the device size is chosen to be close to this number, with a roughly square layout and number of fingers and unit finger width being divisible by powers of two (for ease of scaling). A good fitting pair of numbers is 64 fingers at 12[tm finger width. The finger width was kept low at 12gpm because of electromigration concerns. This device has a total width of 768p.m. 4. 5. Group 1 Aspect Ratio Changes How does the aspect ratio of the layout (number of fingers versus unit finger width) impact the RF performance of the device? The total width of the devices is held constant, so DC they should all behave identical. But device parasitics will differ, as well as the current density in the wiring. 105 This set is also suitable to study electromigration (EM), since the aspect ratio change helps to trade off the current density in the drain and source wires. Devices with many fingers have wide source and narrow drain metal, while devices with high unit finger width have narrow source and wide drain metal. The RF performance may suggest a certain aspect ratio (e.g. square, 1:1) but EM may suggest a different ratio, due to the asymmetric source and drain. We can also find out the maximum device size that EM allows (max. wf determined by EM in source wire, and max nf determined by the accumulation of current in the top drain metal (M4 or higher). This device group includes the devices: id size aspect ratio (NF * pitch / WO) 103 102 101 100 104 105 8 x 96 pan 16 x 48 gm 32 x 24 jun 64 x 12 pun 128 x 6 tm 256 x 3pm 1/46.2 1/11.5 1/2.88 1.39 5.55 22.2 Group 2 Power Cell Scaling Two different approaches of combining multiple cells have been laid out. We wish to answer which style is better performing, and how far scaling of power will be applicable - if we double the number of cells, will we double the power? Do we pay a price in performance (PAE, Linearity)? While theoretically the scaling should be linear, we will have to deal with different devices or cells presenting different impedances to match to - potentially to a point where the performance is degraded because the impedance necessary is out of the range of the matching networks (or load-pull tuners). This group includes 2-D structures and 1-D vertical layout. The vertical layout has less overlap capacitance but also cannot scale to as big devices as the 2-D layout. Each cell is sized NF x WF,G = 64 x 12 jim 1-D Vertical Layout: id 106 107 108 109 110 # cells total width 1 768 gim 2 1536 gm 3 2304 gim 4 3072 gm 6 4608 gm 106 2-D Layout id XbYY 200 201 202 203 204 205 206 207 208 209 210 308 309 310 cells xl x2 x3 x4 x6 xl x2 x3 x4 x6 xl x2 x4 x6 total width 1 768 plm 2 1536 pm 3 2304 jm 4 3072 pm 6 4608 pm 2 1536 pm 4 3072 gm 6 4608 pm 8 6144 pm 12 9216 pm 3 2304 pm 6 4608 pm 12 9216 pm 18 13824 pLm Group 3 Power Cell Sizing Besides trying to study the scaling behavior of power, we are also interested to know how the cell size affects the performance. Issues here are parasitics, electro-migration and self-heating. Starting with the standard cell, we device it up into multiple cells of smaller device size, but overall hold the total width constant. For example, a very large cell may suffer from distributed effects (Rg), self-heating and electromigration. On the other hand, it will have fewer external parasitics. Therefore, there must be an optimum cell size. Devices are sized to give a square aspect ratio. Devices in this group include: id 400 401 402 403 404 405 406 407 size 56 x 13.7 pm 38 x 10.2 prn 32 x 8 pgm 22 x 5.8 pm 16 x 4 pm 14 x 3 Ipm 8x 1.9 mun 4x 1 •m individual widthX by Y # cells 1 lxl 767.2 jLm 2 2x 1 387.6 p.m 3x1 3 256 pLm 3x2 6 127.6 plm 3x4 12 64 gim 3x6 18 42 glm 5 x 10 50 15.2 gim 510x 18 180 4 plm I 10x18 total width 767.2 plm 775.2 gpm 768.0 gim 765.6 gpm 768.0 pm 756.0 jgm 760.0 gLm 720.0 pm The last two layouts (406 and 407) will be useful to study the impact of self-heating and EM, since they should not express these effects due to their small size. However, parasitic capacitances may be very high. 107 Group 4 L, and T,, Variations This set studies the impact of the gate length and oxide thickness (i.e. I/O FETs). How much RF power can each drive? We may support higher voltages at the I/O0 devices, but at the cost of lower ft and fmax. This set is a good indicator of device performance as a function of technology node, since the intrinsic device performance of the 1/O options is close to the older technologies (0.251tm, 0.131im, etc). We are also able to determine if a performance loss/gain is due to the Tox or the Lg increase. Comparing this set of devices to the same I/O devices on the 90 nm wafers can be interesting, since intrinsic device performance should be identical. However, the groundrules have changed, and we can compare EM and performance with the differently sized backend metals in 65 and 90 nm. Devices have constant total width, but aspect ratio adjusted. They include: id 500 501 502 503 504 type DGNFET EGNFET NFET EGNFET NFET size 36 x 21 pm 38 x 20 •m 38 x 20 •m 43 x 18 pm 46 x 16 pm width 756 pm 760 pm 760 pm 774 pm 736 pm Li 0.24 pm 0.24 pm 0.24 pm 0.12 pm 0.12 pm Group 5 Test Structures for Circuits These experiments include a cascade (CS-CG) configuration as well as a grid of 2x2 cells where only certain devices have the input connected to the gate. The other gates are connected to a DC pad for biasing. This should help to study how the PAE behaves at low power levels if we simply switching cells on/off. The performance will be not as good because the switched off devices still contribute their drain parasitics to the output. Can we find a single matching network that provides good performance for the 2x2 cell with 0,1,2 and 3 devices turned off? This is important for applications where we require high PAE at different output power levels. id 302 description cascode, with Vbias of common-gate stage on a separate DC pad. id cell on vs. off 206 4 vertical 4/4 on 304 4 vertical 3/4 on 303 4 vertical 2/4 on 302 4 vertical 1/4 on Devices that are off have their gate connected to the DC pad instead of port 1 on the RF padset. 108 Group 6 Special Devices This group includes devices with separate body contacts and experiments on isolation and EM. Body Contact 300 standard device, but with body contact connected to DC pad 301 same as 300, but body contact at greater distance away from device * study the impact of body voltage on performance, and impact of substrate resistance. Also useful in model development De-embedding 408 de-embedding structure: standard device, OPEN 409 de-embedding structure: standard device, SHORT 410 standard device, no contacts to the drain and source. use for measuring the backend capacitances Isolation 505 standard device, with M1/M2 ground plane under gate and drain. 506 standard device, but with body contact running from top/bottom shielding the gate side of the device from the drain side * study how isolation and ground planes affect the matching and performance Current Flow 507 1 gim of vias and M3 over the source 508 vias and M3 fully over the source 509 vias and M3 fully over the source, and maximum width (factor 2.2x) M2 and M3 source wires * use these to study the tradeoff of increased source current density (by reducing peak current in M2) and parasitic drain-source capacitance Other 510 CA landing on PC aligned with the gate * fewer CAs, but directly centered at the gate fingers. Does this improve performance? In the standard device, PC gate fingers sometimes start in-between two CAs, increasing gate resistance. Group 7 EM Lines These structures are to study the EM behavior of a M1-M2 line under DC and AC+DC current. They consist of a thin metal line on M1 connected to M2 by the number of vias mentioned. 607 608 M1 - 2x V1 - M2 M1 - 5x V1 - M2 609 Ml - 5x VI - M2, with short fins 109 610 Ml - 5x V1 - M2, with long fins The fins in 609 and 610 are used to study how heating affects EM. They are intended as heat sinks. Device Summary id 100 101 102 103 104 105 106 107 108 109 110 200 201 202 203 204 205 206 207 208 209 210 300 301 302 303 304 305 Grcaup(s) device size 1 64 x 12 1 32 x 24 1 16 x 48 1 8 x 96 1 128 x 6 1 256 x 3 2 64 x 12 2 64 x 12 2 64 x 12 2,5 64 x 12 2 64 x 12 2 64 x 12 2 64 x 12 2 64 x 12 2 64 x 12 2 64 x 12 2 64 x 12 2 64 x 12 2 64 x 12 2 64 x 12 2 64 x 12 2 64x 12 6 64 x 12 6 64 x 12 5 64 x 12 5 64 x 12 5 64 x 12 5 64 x 12 306,307 - do not exist 308 2 64 x 12 309 2 64 x 12 310 2 64x12 400 3 56 x 13.7 401 3 38 x 10.2 402 3 32 x 8 403 3 22 x 5.8 404 3 16 x 4 c4~11(s) specials standard cell 1 vertical 2 vertical 3 vertical 4 vertical 6 vertical 1 xl 1 x2 1 x3 1 x4 1 x6 2 xl 2 x2 2 x3 2 x4 2 x6 3 xl body contact (standard) body contact (far) 4 vertical 1 device on, 3 connected to DC pad 4 vertical 2 device on, 2 connected to DC pad 4 vertical 3 device on, 1 connected to DC pad 2 in cascode cascade (CS-CG) with CG's gate at DC pad 3 x2 3 x4 3 x6 1 xl 2 xl 3 xl 3 x2 3 x4 110 405 3 406 3 407 3 408 6 409 6 410 6 500 4 501 4 502 4 503 4 504 4 505 6 506 6 507 6 508 6 509 6 510 6 600-606 - do not 607 7 608 7 609 7 610 7 14 x 3 3x6 8 x 1.9 5x10 10 x 18 4x 1 64 x 12 64 x 12 64 x 12 36 x 21 38 x 20 38 x 20 43 x 18 46 x 16 64 x 12 64 x 12 64 x 12 64 x 12 64 x 12 64 x 12 OPEN SHORT No Source/Drain CA DGNFET, LG=0.24um EGNFET, LG=0.24um NFET, LG=0.24um EGNFET, LG=0.12um NFET, LG=0.12um Ml ground plane body contact isolation part M3 source wire over M2 full M3 source wire over M2 M2/M3 source wire at 2.2x width PC/CA aligned exist 2 vias 5 vias 5 vias, short fin 5 vias, long fin Device ILocation in Macros There are three macros (ac_pwrl-3) and one long horizontal macro with two device rows. The devices are found: AC PWR1 -other-other-other-other-other-other510 100 101 102 103 AC PWR2 AC PWR3 200 500 201 501 202 502 203 503 204 504 205 505 206 506 207 104 208 105 209 408 210 409 AC_PwrHoriz (no label b/c of space issues) 400 401 402 403 404 405 406 407 308 309 410 310 300 301 302 303 304 305 106 107 108 109 110 507 508 509 The 607-610 macro is found on the side of the block containing the above horizontal macro. 111 90 nm Devices Devices on the 90 nm wafer are grouped the same way as previously the 65 nm devices in 2.1.1. Only new aspects compared to the ones mentioned above in the 65 nm section are shown below: Group 1 Aspect Ratio Changes id size aspect ratio (nf * pitch / wf) 102 12 x 64 jim 1/14.8 101 100 103 104 24 x 32 gm 48 x 16 Lm 96 x 8 gim 192 x 4 gm 1/3.7 1.08 4.32 17.28 Group 2 Power Cell Scaling 1-D Vertical Layout: id # cells total width 200 201 202 2 3 4 1536 ptm 2304 gtm 3072 gLm 2-D Layout id X by Y 109 203 204 205 1x2 2 x1 2x2 2x4 # cells total width 2 2 4 8 1536 gtm 1536 gm 3072 gim 6144 jlm Group 3 Power Cell Sizing id 105 106 107 206 207 size 32 x 12 pm 24 x 8 jm 16 x 6 jm 12 x 16 pm 24 x 16 gm individual widthX by Y # cells 384 im 2 x1 2 192 gm 96 gm 192 jim 384 gm 2x2 2x4 2x 4 2x2 4 8 8 4 112 total width 768 jLm 768 jim 768 jim 1536 ptm 1536 jlm 208 209 203 210 96 x 16 [Lm 48 x 8 jm 48 x 16 gm 48 x 32 Wmn 1536 jim 384 jim 768 jlm 1536 gim 1536 gim 1536 gm 1536 gm 1536 gm 2x2 2x1 206-210 are used to study the tradeoff of expanding the cell along the number of fingers or unit finger width axis (device size and id number in parenthesis): 48x8 (209) 12x16 24x16 48x16 96x16 (206) (207) (203) (208) 48x32 (210) Alternatively, instead of 203 for the 48x16 center, 109 can also be used (the difference is the way the two 48x16 cells are arranged). Group 4 Lg and T.x Variations id 400 401 402 403 404 ---- tvDe m NFET NFET EGNFET EGNFET DGNFET width 544 jm x 20 11m 600 jlm x 16 9im 544 jim x 20 jim 600 jm x 20 9lm 600 gm size x 16 jim 0.12 0.24 0.12 0.24 0.24 jlm jlm jlm jlm Group 5 Test Structures for Circuits There are no structures of this group in 90 nm. Group 6 Special Devices This group includes devices with separate body contacts and experiments on isolation and EM. Body Contact 300 standard device, but with body contact connected to DC pad 301 same as 300, but body contact at greater distance away from device 113 De-embedding 304 de-embedding structure: standard device, OPEN 305 de-embedding structure: standard device, SHORT 110 de-embedding structure: standard device, THROUGH Isolation These devices should help to study how much the ground plane shielding impacts efficiency (PAE). 305 standard device, with M1/M2 ground plane under gate and drain. 311 standard device, but without any ground plane under the gate and drain pads Maximum Width To study the behavior of cell vs. single device for very large devices, as well as the power limits of CMOS. id 308 309 size cell width 192 x 64 pm 12288 gm 64 x 24 pm 1536 gm cell 1 2x4 total width 12288 gtm 12288 gtm Capacitance and EM Can widening the MI source metal improve performance by reducing gate-to-drain feedback capacitance? What is the impact of increasing the gate pitch and widening the metal M1 and M2 going to the source? 308 G/S overlap: widen M1 as much as groundrules permit 307 increase the spacing of the gate, so that the metal can be thicker 109 flipped source and drain routing into the device. Source comes in vertical now. Helps in EM investigation and study of parasitics. Other 303 Layout using the standard RF PCell. Answers the question how much improvement in performance and EM reliability the layouts have brought compared to a simple (naive?) layout. Group 7 EM Lines There are no structures of this group in Tomcat. 114 Device Summary Summary device size cell(s) Group(s) Device 100 ,48 x 16 id 101 102 103 104 105 106 107 108 109 110 200 201 202 203 204 205 206 207 208 209 210 300 301 302 303 304 305 306 307 308 309 310 400 401 402 403 404 specials standard cell 24 x 32 12 x 64 96 x 8 192x 4 32 x 12 24 x 8 16 x 6 48 x 16 2x1 2x2 2x4 1x2 flip s/d through 48 x 16 48x 16 48 x 16 48 x 16 48 x 16 48 x 16 48 x 16 48 x 16 12 x 16 24 x 16 96x 16 48 x 8 48 x 32 48 x 16 48 x 16 48 x 16 48x 16 48 x 16 48 x 16 48 x 16 48x 16 192 x 64 64 x 24 48 x 16 34 x 16 30 x 20 34x 16 30 x 20 30 x 20 2 verti cal 3 verti cal 4 vertical 2x1 2x2 2x4 2x4 2x2 2x2 2x4 separate body contact separate body contact, far from device IBM RF PCell open short ground plane increased metal 1 and 2, gate pitch G/S overlap on M1 maximum width maximum width in cells no ground plane under pads NFET, Lg = 0.12um NFET, Lg = 0.24um EGNFET, Lg = 0.12um EGNFET, Lg = 0.24um DGNFET, Lg = 0.24um 115 Device Location in Macros There are four macros (ac_pwrl-4) The devices are arranged as follows: AC_PWR1 AC PWR2 AC_PWR3 AC PWR4 100 200 300 400 101 201 301 401 102 202 109 402 103 203 410** 403 104 204 310 404 105 205 305 405** 306 406** 106 206 107 207 307 407** 108 208 308 305*** 302 209 309 409** 110 210 304 303 ** non-functional, as source and drain are shorted together *** wrong device 116 Bibliography [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] J. Scholvin, D.R. Greenberg, J.A. del Alamo, "RF power potential of 90 nm CMOS: device options, performance, and reliability", IEEE International Electron Devices Meeting, Dec 2004, pp. 4 5 5 -4 5 8 . W. Schuchter et al., IEEE RFIC, June 2001, pp. 183-186 I. Aoki et al., IEEE CICC, May 2001, pp. 57-60 Y Zhang et al., IEEE RFIC, June 2004, pp. 563-566 D. Su et al., IEEE CICC, May 1997, pp. 189-192 H.M. Hsu et al., IEEE Trans. MTT, Dec 2002, pp. 2873-2881 M. Ferndahl, H.O. Vickes, H. Zirath, I. Angelov, F. Ingvarson, A. Litwin; "90-nm CMOS for microwave power applications", Microwave and Wireless Components Letters, IEEE, Volume: 13, Issue: 12, Dec. 2003, pp. 523 - 525. D. James, "2004- The Year of 90-nm: A Review of 90 nm Devices", IEEE/SEMI Advanced Semiconductor Manufacturing Conference and Workshop, Apr. 2005, pp. 7276. RF2681, preliminary product data sheet, RF Micro Devices, Rev A8 040512, pp. 529 536 E. Chen, D. Heo, M Hamai, J. Laskar, D. Bien; "0.24-um CMOS technology for Bluetooth power applications", IEEE Radio and Wireless Conference, 2000, pp. 163 166. J. Scholvin, D.R. Greenberg, and J.A. del Alamo, "Performance and limitations of 65 nm CMOS for integrated RF power applications", IEEE Int. Electron Devices Meeting, 2005, pp. 369-372 W. Jeamsaksiri, D. Linten, S. Thijs, G. Carchon, J. Ramos, A. Mercha, X. Sun, P. Soussan, M. Dehan, T. Chiarella, R. Venegas, V. Subramanian, A. Scholten, P. Wambacq, R. Velghe, G. Mannaert, N. Heylen, R. Verbeeck, W. Boullart, I. Heyvaert, M. I. Natarajan, G. Groeseneken, I. Debusschere, S. Biesemans, S. Decoutere, "A lowcost 90nm RF-CMOS platform for record RF circuit performance", IEEE 2005 Symposium on VLSI Technology, pp. 60 - 61. L.F. Tiemeijer, R.J. Havens, R. De Kort, A.J. Scholten, R. Van Langevelde, D.B.M. Klaassen, G.T. Sasse, Y. Bouttement, C. Petot, S. Bardy, D. Gloria, P. Scheer, S. Boret, B. Van Haaren, C. Clement, J.F. Larchanche, I.S. Lim, A. Zlotnicka, A. Duvallet, "Record RF performance of standard 90 nm CMOS technology", IEEE International Electron Devices Meeting, Dec 2004, pp. 441-444. B. Jagannathan, R. Groves, D. Goren, B. Floyd, D. Greenberg, L. Wagner, S. Csutak, S. Lee, D. Coolbaugh, J. Pekarik, "RF CMOS for Microwave and MM-wave Applications", IEEE 2006 Topical Meeting on Silicon Monolithic Integrated Circuits in RF Systems, pp. 259-264. P.H. Woerlee, M.J. Knitel, R. van Langevelde, D.B.M. Klaassen, L.F. Tiemeijer, A.J. Scholten, A.T.A. Zegers-van Duijnhoven, "RF-CMOS Performance Trends", IEEE Transactions on Electron Devices, Vol. 48, No. 8, Aug 2001, pp. 1776-1782. IBM 65 nm press coverage http://www.eet.com/news/semi/showA rticle.jhtml?articlelD=164302481 117 [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] E. Alekseev, et al, Trans. MTT, Oct 2000, pp. 1694-1700 P. Kurpas, et al, IEDM 2002, pp. 681-684 F. van Raay, et al, MTT-S 2003, pp. 451-454 H. Brech, et al, MTT-S 2003, pp. 209-212 I.P. Smorchkova, et al, Trans. MTT, Feb 2003, pp. 665-668 E. Ferro, F Potorti, "Bluetooth and Wi-Fi Wireless Protocols: A Survey and A Comparison", IEEE Wireless Communications, Feb. 2005, pp. 12-26. B. Jagannathan, R. Groves, D. Goren, B. Floyd, D. Greenberg, L. Wagner, S. Csutak, S.J. Lee, D. Coolbaugh, J. Pekarik, "RF CMOS for microwave and MM-wave applications", IEEE Topical Meeting on Silicon Monolithic Integrated Circuits in RF Systems, 2006, pp. 259-264. C.S. Chang, C.P. Chao, G.J. Chern, Y.C. Sun, "Advanced CMOS Technology Portfolio for RF IC Applications", IEEE Transactions on Electron Devices, Vol. 52, No. 7, July 2005, pp. 1324-1334. D. Buss, B.L. Evans, J. Bellay, W. Krenik, B. Haroun, D. Leipold, K. Maggio, J.Y. Yang, T. Moise, "SOC CMOS Technology for Personal Internet Products", IEEE Transactions on Electron Devices, Vol. 50, No. 3, March 2003, pp. 5 4 6 -5 5 6 UC Berkeley BSIM documentation: http://www-device.eecs.berkeley.edul~-bsim31 Bond Wire C. Yoo, Q. Huang, "A Common-Gate Switched 0.9-W Class-E Power Amplifier with 41% PAE in 0.25-ym CMOS", IEEE Journal of Solid State Circuits, Vol. 36, No. 5, May 2001, pp. 823-830. T.C. Kuo, B.B. Lusignan, "A 1.5W Class-F RF Power Amplifier in 0.2 jtm CMOS Technology", ISSCC 2001, pp. 154 - 15 5 . S.C. Cripps, "RF Power Amplifiers for Wireless Communications", Artech House, 1st Edition, 1999, pp. 54-60 The following is a list of references used in compiling the power data in chapter 1. The references are abbreviated to save space and only mention the authors, publication and year. [31] SP Knight, DA Daly, M Caulton ; IEEE ISSCC; 1/1968 [32] PT Chen, HP; IEEE ISSCC; 1/1973 [33] [34] L Napoli, R DeBrecht ; IEEE MTT-S; 6/1973 PT Chen ; IEEE J SS Circ; 7/1974 [35] [36] YS Wu, HT Yuan, JB Kruger ; IEEE ISSCC; 1/1976 HQ Tserng, V Sokolov, HM Macksey, WR Wissman ; IEEE MTT-S; 6/1976 [37] Y Arai, T Kouno, T Horimatsu, H Komizo ; IEEE Trans MTT; 6/1976 [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] JG Oakes, RA Wickstrom, DA Tremere, TMS Heng ; IEEE Trans MTT; 6/1976 M Fukuta, K Suyama, H Suzuki, Y Nakayama, H Ishikawa ; IEEE Trans MTT; 6/1976 Y Saito, S Akasaka, S Fukuda, I Haga ; IEEE MTT-S; 6/1978 HT Yuan, YS Wu ; IEEE MTT-S; 6/1978 PT Ho ; IEEE MTT-S; 6/1978 KB Niclas, RB Gold, WT Wilser, WR Hitchens ; IEEE J SS Circ; 7/1978 V Sokolov, RE Williams, DW Shaw ; IEEE ISSCC; 1/1979 PT Ho, CM Pham, RL Mencik ; IEEE MTT-S; 4/1979 RA Pucel, P Ng, J Vorhaus ; IEEE MTT-S; 4/1979 S Aihara, Y Fujiki, S Fukuda, S Akasaka, I Haga ; IEEE MTT-S; 4/1979 H Abe, Y Aono ; IEEE Trans MTT; 4/1979 JTC Chen, CP Snapp ; IEEE Trans MTT; 4/1979 118 [50] [51] [52] [53] [54] [55] [56] [57] [58] [59] [60] [61] [62] [63] [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85] [86] [87] [88] [89] [90] [91] [92] [93] [94] [95] [96] [97] [98] [99] [100] [101] [102] [103] Y Kadowaki, M Nakatani, T Ishii, S Mitsui, K Shirahata ; IEEE MTT-S; 12/1979 M Nakatani, Y Kadowaki, T Ishii ; IEEE Trans MTT; 12/1979 JA Higgins, RL Kuvas ; IEEE Trans MTT; 1/1980 KB Niclas, WT Wilser, RB Gold, WR Hitchens; IEEE Trans MTT; 3/1980 JL Vorhaus, RA Pucel, Y Tajima, W Fabian ; IEEE ISSCC; 1/1981 Y Mitsui, M Kobiki, M Wataze, K Segawa, M Otsubu, T Ishii ; IEEE Trans MTT; 4/1981 J Sone, Y Takayama ; IEEE Trans MTT; 4/1981 D Laighton, J Sasonoff, J Selin ; IEEE MTT-S; 6/1981 RL Camisa, JB Klatskin, A Mikelsons ; IEEE MTT-S; 6/1981 V Sokolov, RC Bennett; IEEE MTT-S; 6/1981 SR Mazumder, T Dao, TL Tsai, WC Tsai ; IEEE MTT-S; 6/1982 J Goel, G Oransky, S Yuan, P O'Sullivan, J Burch; IEEE MTT-S; 6/1982 T Tsukii, Y Tajima, PD Miller, R Mozzi, EK Tong ; IEEE ISSCC; 1/1983 A Barzeghi, F Oggionni, G Pratesi ; IEEE MTT-S; 6/1983 F Sechi, R Brown, P Jozwiak, G Rolland, J Brown, E Mykietyn; IEEE MTT-S; 6/1983 RG Gels, RD Standley, DL Wilson, R Trambarulo ; IEEE MTT-S; 6/1983 T Noguchi, Y Aono, K Watanabe, S Kameda ; IEEE J Sel A Com; 9/1983 Y Tokumitsu, T Saito, N Okubo, Y Kaneko; IEEE Trans MTT; 3/1984 BM Kim, HQ Tserng, P Saunier ; IEEE Trans MTT; 3/1984 Y Ayasli, LD Reynolds, RL Mozzi, LK Hanes ; IEEE Trans MTT; 3/1984 Y Hirachi, Y Takeuchi, M Igarashi, K Kosemura, S Yamamoto ; IEEE Trans MTT; 3/1984 CD Palmer, P Saunier, RE Williams; IEEE MMWMC; 4/1984 CD Chang, SK Wang, LCT Liu, M Siracusa, H Yamasaki, JM Schellenberg ; IEEE MMWMC; 4/1985 P Saunier, HQ Tserng, B Kim, GH Westphal ; IEEE MMWMC; 6/1985 HM Macksey, HQ Tserng, HD Shih ; IEEE MMWMC; 6/1985 DM Drury, DC Zimmermann, DE Zimmerman ; IEEE MTT-S; 6/1985 A Ezzeddine, HLA Hung HC Huang ; IEEE MTT-S; 6/1985 AAM Saleh, MF Wazowicz ; IEEE Trans MTT; 7/1985 E Belohoubek, R Brown, H Johnson, A Fathy, D Bechtle, D Kalokitis, E Mykietyn ; IEEE MTT-S; 6/1986 SK Wang, KG Wang, CD Chang; IEEE MMWMC; 6/1986 JR Lane, RG Freitag, JE Degenford, M Cohn ; IEEE MTT-S; 6/1986 KB Bhasin, IDJ Connolly ; IEEE Trans MTT; 10/1986 D Bechtle, J Klatskin, G Taylor, M Eron, SG Liu, RL Camisa, H Dudley ; IEEE MTT-S; 6/1987 S Nishiki, T Nojima ; IEEE MTT-S; 6/1987 HLA Hung, A Ezzeddine, LB Holdeman, FR Phelleps, JF Allison, AB Cornfeld, T Smith, HC Huang; IEEE MMWMC; 6/1987 JJ Komiak; IEEE GaAs IC; 1/1988 E Nakanishi, J Shibata, O Ohsawa; IEEE; 1/1988 AE Geissberger, IJ Bahl, EL Griffin, RA Sadler ; IEEE TED; 4/1988 BD Geller, PE Goettle ; IEEE MTT-S; 4/1988 B Kopp, DD Heston ; IEEE MTT-S; 4/1988 M Avasarala, DS Day, S Chan, P Gregory, JR Basset ; IEEE MTT-S; 4/1988 T Nojima, S Nishiki ; IEEE MTT-S; 4/1988 Y Oda, S Arai, T Yoshida, H Nakamura, S Yanagawa, S Hori, K Kamei ; IEEE MTT-S; 6/1988 C Peignet, Y Mancusco, G LeMeur, L Remiro, A Bert, JF Jouen, P Savary ; IEEE MTT-S; 6/1988 SD Pritchett ; IEEE MMWMC; 6/1988 G Hegazi, HLA Hung, JL Singer, FR Phelleps, AB Cornfeld, T Smith, JF Bass, HE Carlson, HC Huang; IEEE MMWMC; 6/1988 N Camilleri, B Kim, HQ Tserng, HD Shih ; IEEE MMWMCS; 6/1988 RB Culbertson, DC Zimmermann; IEEE; 6/1988 JA Higgins ; IEEE GaAs IC; 10/1988 P Saunier, HQ Tserng, N Camilleri, K Bradshaw, HD Smith ; IEEE GaAs IC; 10/1988 HLA Hung, G Hegazi, KE Peterson, FR Phelleps, HC Huang ; IEEE GaAs IC; 10/1988 MJ Schindler, JP Wendler, MP Zaitlin, ME Miller, JR Dormail ; IEEE Trans MTT; 12/1988 KH Snow, JJ Komiak, DA Bates ; IEEE Trans MTT; 12/1988 HLA Hung, GM Hegazi, TT Lee, FR Phelleps, JL Singer, HC Huang ; IEEE Trans MTT; 12/1988 119 [104] [105] [106] [107] [108] [109] [110] [111] JL Vorhaus ; IEEE MTT-S; 6/1989 KE Peterson, HL Hung, FR Phelleps, TF Noble, HC Huang ; IEEE MTT-S; 6/1989 IJ Bahl, R Wang, A Geissberger, E Griffin; IEEE MMWMCS; 6/1989 M Avasarala, DS Day, S Chan, C Hua, JR Basset ; IEEE MMWMCS; 6/1989 Y Oda, T Yoshida, K Kai, S Arai, S Yanagawa; IEEE MMWMCS; 6/1989 AL Riley, D Rascoe, V Lubecke, J Huang, L Duffy ; IEEE MMWMCS; 6/1989 FS Auricchio, RA Rhodes, DS Day; IEEE MTT-S; 6/1989 DW Ferguson, PM Smith, PC Chao, LF Lester, RP Smith, P Ho, A Jabra, JM Ballingall ; IEEE MTT-S; 6/1989 [112] YC Shih, K Tan, K Kasel, KK Yu, SK Wang ; IEEE MTT-S; 6/1989 [113] PM Smith, MY Kao, P Ho, PC Chao, KHG Duh, AA Jabra, RP Smith, JM Ballingall ; IEEE MTT-S; 6/1989 [114] AP Long, IH Goodridge, JP King, AJ Holden, JG Metcalfe, RF Hayes, R Nicklin, RC Goodfellow ; IEEE Elec Letters; 7/1989 [115] S Sriram, RC Clarke, RL Messham, TJ Smith, MC Driver ; IEEE HSSDC; 7/1989 [116] IJ Bahl, EL Griffin, AE Geissberger, C Andricos, TF Brukiewa ; IEEE Trans MTT; 9/1989 [117] N Camilleri, P Chye, R Prioriello ; IEEE GaAs IC; 10/1989 [118] M Avasarala, DS Day, C Hua, S Chan, JR Basset ; IEEE GaAs IC; 10/1989 [119] B Bayraktaroglu, MA Khatibzadeh, RD Hudgens; IEEE GaAs IC; 10/1989 [120] AW Swanson, PM Smith, PC Chao, KH Duh, JM Ballingall ; IEEE; 12/1989 [121] VD Hwang, YC Shih, TY Chi, DC Wang ; IEEE; 1/1990 [122] B Bayraktaroglu, MA Khatibzadeh, RD Hudgens ; IEEE; 1/1990 [123] AK Ezzeddine, HLA Hung, HC Huang ; IEEE Trans MTT; 4/1990 [124] M Minowa, M Onoda, E Fukuda, Y Daido ; IEEE Veh Tech Conf; 4/1990 [125] MA Khatibzadeh, HQ Tserng ; IEEE ; 4/1990 [126] MA Khatibzadeh, B Bayraktaroglu ; IEEE; 4/1990 [127] RD Boesch, JA Thompson ; IEEE ; 6/1990 [128] T Ho, F Phelleps, G Hegazi, K Pande, H Huang, P Rice, P Pages; IEEE MILCOM; 10/1990 [129] B Geller, J Tyler, L Holdeman, F Phelleps, P Cline, G Laird ; IEEE GaAs IC; 10/1990 [130] H Le, YC Shih, T Chi, K Kasel, DC Wang; IEEE ; 10/1990 S Sriram, RL Messham, TJ Smith, HM Hobgood, MC Driver ; IEEE ; 10/1990 [131] [132] SR LeSage, MJ Mendes, MJ Sullivan ; IEEE; 10/1990 [133] R Ramachandran, M Nijar, A Podell, E Stoneham, S Mitchell, NL Wang, WJ Ho, MF Chang, GJ Sullivan, JA Higgins, PM Asbeck, T Cisco ; IEEE; 10/1990 [134] M Avasarala, DS Day, S Chan, C Hua, JR Basset ; IEEE; 10/1990 [135] JJ Komiak ; IEEE Trans MTT; 12/1990 [136] DC Streit, KL Tan, RM Dia, JK Liu, AC Han, JR Velebir, SK Wang, TQ Trinh, PMD Chow ; IEEE EDL; 4/1991 [137] DT Bryant ; IEEE MTT-S; 6/1991 [138] JA Lester, WL Jones, PD Chow ; IEEE MTT-S; 6/1991 [139] G Hegazi, E Chang, J Singer, F Phelleps, P McNally, K Pande ; IEEE MTT-S; 6/1991 [140] H Takeuchi, M Muraoka, T Hatakeyama, A Matsuoka, M Honjou, S Miyazaki, K Tanaka, T Nakata ; IEEE MTT-S; 6/1991 [141] J Ozaki, K Arai, M Miyauchi, S Watanabe, S Kamihashi ; IEEE MMWMCS; 6/1991 A Asensio, F Perez; IEEE ISSCC; 6/1991 [142] [143] H Le, YC Shih, V Hwang, T Chi, K Kasel, DC Wang; IEEE; 6/1991 [144] M Gat, DS Day, JR Basset; IEEE; 6/1991 [145] DW Ferguson, SA Allen, MY Kao, PM Smith, PC Chao, MAG Upton, JM Ballingall ; IEEE ; 6/1991 [146] FW Smith, CL Chen, LJ Mahoney, MJ Manfra, DH Temme, BJ Clifton, AR Calawa ; IEEE ; 6/1991 [147] D Helms, JJ Komiak, WF Kopp, P Ho, PM Smith, RP Smith, D Hogue ; IEEE ; 6/1991 [148] KE Peterson, HLA Hung, FR Phelleps, EY Chang, JL Singer, HE Carlson, AB Cornfeld ; IEEE ; 6/1991 [149] M Gat, DS Day, JR Basset ; IEEE ; 6/1991 [150] B Maoz, BE Bedard, A Oki ; IEEE BCTM; 9/1991 [151] HM Le, YC Shih, VD Hwang, TY Chi, KJ Kasel, DC Wang ; IEEE; 10/1991 [152] P Bartusiak, T Henderson, T Kim, A Khatibzadeh, B Bayraktaroglu ; IEEE Elec Letters; 10/1991 120 [153] [154] [155] [156] [157] [158] [159] [160] [161] [162] [163] [164] [165] [166] [167] [168] [169] [170] [171] [172] [173] [174] [175] [176] [177] [178] [179] [180] [181] [182] [183] [184] [185] [186] [187] [188] [189] [190] [191] [192] [193] [194] [195] [196] [197] [198] [199] [200] H Yang, RM Herman, KW Angel, AM Chao, MJ Schindler, M Adlerstein, Y Tajima, DM Danzilio ; IEEE GaAs IC; 10/1991 M Gat, DS Day, S Chan, C Hua, JR Basset ; IEEE ; 10/1991 HM Le, YC Shih, T Chi, L Fong, K Kasel ; IEEE; 10/1991 TC Cheng ; IEEE IEE Coil on SSPA; 12/1991 D Ryan, G Railton ; IEEE IEE Coll on SSPA; 12/1991 GJ Ayling, JS Joshi ; IEEE IEE Coil on SSPA; 12/1991 G Collinson, CW Suckling ; IEEE IEE Coll on SSPA; 12/1991 G Hegazi, K Pande, F Phelleps, E Chang, A Cornfeld, P Rice, M Ghahremani, P Pages ; IEEE MWGW Letters; 1/1992 J Mondal, J Geddes, D Carlson, M Vickberg, S Bounnak, C Anderson ; IEEE ; 3/1992 T Takagi, Y Ikeda, K Seino, G Toyoshima, A Inoue, N Kasai, M Takada ; IEEE MMWMCS; 6/1992 MA Khatibzadeh, B Bayraktaroglu, T Kim; IEEE MMWMCS; 6/1992 M Cicolani ;IEEE MTT-S; 6/1992 TH Miers, VA Hirsch ; IEEE MTT-S; 6/1992 J Goel, KL Tan, DI Stones, RW Chan, DC Streit, S Peratoner, J Schellenberg ; IEEE MTT-S; 6/1992 D Raicu, BM Kraemer, DS Day, JR Basset, J Wei, C Hua, Y Chung, CS Chang; IEEE MTT-S; 6/1992 K Johnson, A Lum, S Nelson, E Reese, K Salzman ; IEEE ; 6/1992 HQ Tserng, P Saunier, TC Kao ; IEEE; 6/1992 R Actis, KB Nichols, RW Chick, RA McMorran ; IEEE; 6/1992 JM Schellenberg, KL Tan, RW Chan, CH Chen, TS Lin, DC Streit, PH Liu ; IEEE ; 6/1992 GS Dow, K Tan, N Ton, J Abell, M Siddiqui, B Gorospe, D Streit, P Liu, M Sholley ; IEEE; 6/1992 M Easton, R Basset, DS Day, C Hun, CS Chang, J Wei ; IEEE ; 6/1992 T Ho, G Metze, A Cornfeld, T Lee, K Pande, H Huang, P Ferguson, M Foisy ; IEEE MWGW Letters; 7/1992 JI Upshur, PE Goettle, BD Geller, RK Gupta, F Phelleps ; IEEE Midwest CSC; 7/1992 JP Mondal, JJ Geddes, RC Becker, A Contolatis, M Vickberg, D Carlson, SS Bounnak, C Anderson, V Sokolov; IEEE J SS Circ; 10/1992 H Wang, DC Yang, MV Aust, EA Rezek, BR Allen, LA Fletcher, RC Becker ; IEEE J SS Circ; 10/1992 WJ Soo Hoo, RF Tenney, G Besenyei, PL Ritchie, DR Daugherty, AE Buchanan ; IEEE MILCOM; 10/1992 A Platzker, KT Hetzler, J Bradford Cole ; IEEE ; 10/1992 CK Pao, GL Lan, CS Wu, A Igawa, M Hu, JC Chen, YC Smith ; IEEE ; 10/1992 T Ho, K Pande, F Phelleps, J Singer, P Rice, J Adair, M Ghahremani ; IEEE Elec Letters; 10/1992 TH Chen, KL Tan, GS Dow, H Wang, KW Chang, TN Ton, B Allen, J Berenz, PH Liu, D Streit, G Hayashibara ; IEEE GaAs IC; 10/1992 PJ Bartusiak, T Henderson, T Kim, B Bayraktaroglu ; IEEE; 10/1992 T Arell, TK Hongsmatip ; IEEE; 10/1992 M Goldfarb, B Erfany, P Bernkopf ; IEEE; 10/1992 JJ Komiak, D Helms ; IEEE; 10/1992 S Cooper, K Anderson, K Salzman, R Culbertson, J Mason, D Bryant, P Saunier ; IEEE; 10/1992 S Murai, T Sawai, T Yamaguchi, S Matsushita, Y Harada ; IEEE; 10/1992 WJ Ho, NL Wang, MF Chang, JA Higgins ; IEEE ; 10/1992 NL Wang, WJ Ho, JA Higgins ; IEEE; 10/1992 GW Wroblewski, F McMahon, M Kleidermacher, DW Ferguson ; IEEE; 12/1992 NL Wang, WJ Ho, AL Sailer, JA Higgins ; IEEE MWGW Letters; 3/1993 M Gillick, S Lucyszyn, U Karacaoglu, ID Robertson, AH Aghvami ; IEEE RAMSSSSA; 3/1993 M Matloubian, AS Brown, LD Nguyen, MA Meledes, LE Larson, MJ Delaney, MA Thompson, RA Rhodes, JE Pence ; IEEE ; 4/1993 D Ngo, B Beckwith, P O'Neil, N Camilleri ; IEEE MTT-S; 6/1993 JW Gipprich, LE Dickens, JA Faulkner; IEEE MTT-S; 6/1993 N Camilleri, J Costa, D Lovelace, D Ngo ; IEEE MTT-S; 6/1993 K Kurdoghlian, CS Wu, W Yau, J Chen, M Hu, C Pao, D Bosch; IEEE; 6/1993 MV Aust, B Allen, GS Dow, R Kasody, G Luong, M Biedenbender, K Tan; IEEE; 6/1993 S Hara, H Sato, JK Twynam, M Akagi, N Nambu, N Tanba, K Yoshikawa, T Kinosada, M Yagura, H Tsuji, T Shinozaki, T Yoshimasu, T Miyajima, T Tomita; IEEE; 6/1993 121 [201] [202] [203] [204] [205] [206] [207] [208] [209] [210] [211] [212] [213] [214] [215] [216] [217] [218] [219] [220] [221] [222] [223] [224] [225] [226] [227] [228] [229] [230] [231] [232] [233] [234] [235] [236] [237] [238] [239] [240] [241] [242] [243] [244] JJ Komiak, LW Yang, RS Brozovich, ST Fu, DP Smith; IEEE; 6/1993 R Yarborough, P Saunier, HQ Tserng, K Salzman, B Smith, D Heston; IEEE; 6/1993 CH Chen, HC Yen, K Tan, L Callejo, G Onak, DC Streit, PH Liu, JM Schellenberg ; IEEE ; 6/1993 DT Bryant, K Salzman, R Hudgens ; IEEE; 6/1993 M Muraguchi, M Nakatsugawa, M Aikawa ; IEEE ; 6/1993 C Duvanaud, P Bouysse, JM Nebus, L Lapierre, JP Villotte ; IEEE ; 6/1993 S Toyoda ; IEEE ; 6/1993 H Wang, C Pinatel, M Tsouli, JO Plouchart, A Konczykowaska, M Riet, S Vuye, P Berdaguer ; IEEE ; 6/1993 HQ Tserng, P Saunier ; IEEE ; 6/1993 RN Simons, RQ Lee ; IEEE; 6/1993 M Mochizuki, Y Itoh, M Kohno, H Masuno, T Takagi; IEEE; 6/1993 M Matloubian, L Larson, A Brown, L Jelloian, L Nguyen, M Lui, T Lui, J Brown, M Thompson, W Lam, A Kurdoghlian, R Rhodes, M Delaney, J Pence; IEEE; 6/1993 LE Larson, MM Matloubian, JJ Brown, AS Brown, R Rhodes, D Crampton, M Thompson ; IEEE; 7/1993 F Ali, M Salib, A Gupta; IEEE ; 7/1993 C Duvanaud, S Dietsche, G Pataut, J Obregon ; IEEE; 7/1993 M Salib, A Gupta, F Ali, D Dawson ; IEEE; 9/1993 FJ McGrath, P O'Sullivan, E Heaney, G Dawe, G St Onge, K Jackson, C Kermarrec, B Fahey ; IEEE WESCON; 9/1993 CK Pao, WS Wong, WD Gray, C Liu, DC Wang, CP Wen ; IEEE EPEP; 10/1993 T Arell, TK Hongsmatip ; IEEE; 10/1993 B Hughes, J Orr, G Martin; IEEE GaAs IC; 10/1993 F Ali, M Salib, A Gupta, D Dawson ; IEEE GaAs IC; 10/1993 MD Biedenbender, JL Lee, KL Tan, PH Liu, A Freudenthal, DC Streit, G Luong, R Lai, MV Aust, B Allen, TS Liu, HC Yen; IEEE GaAs IC; 10/1993 H Sato, M Miyauchi, K Sakuno, M Akagi, M Hasegawa, JK Twynam, K Yamamura, T Tomita; IEEE GaAs IC; 10/1993 M Funabashi, K Hosoya, K Ohata, K Onda, N Iwata, M Kuzuhara; IEEE GaAs IC; 10/1993 A Kurdoghlian, W Lam, C Chou, L Jellian, A Igawa, M Matloubian, L Larson, A Brown, M Thompson, C Ngo ; IEEE GaAs IC; 10/1993 D Deakin, WJ Ho, EA Sovero, J Higgins; IEEE GaAs IC; 10/1993 ST Fu, JJ Komiak, LF Lester, KHG Duh, PM Smith, PC Chao, TH Yu ; IEEE GaAs IC; 10/1993 JC Huang, W Boulais, A Platzker, T Kazior, L Aucoin, S Shanfield, A Bertrand, M Vafides, M Niedzwiecki ; IEEE GaAs IC; 10/1993 P O'Sullivan, G St Onge, E Heaney, F McGrath, C Kermarree ; IEEE GaAs IC; 10/1993 W Lam, M Matloubian, A Kurdoghlian, L Larson, A Igawa, C Chou, L Jelloian, A Brown, M Thompson, C Ngo; IEEE; 10/1993 S D'Agostino, C Paoloni ; IEEE; 10/1993 M Hanczor, M Kumar ; IEEE Trans MTT; 12/1993 TC Ho, SW Chen, K Pande, PD Rice; IEEE Trans MTT; 12/1993 T Yoshimasu, N Tanba, S Hara; IEEE; 3/1994 Y Ota, C Adachi, H Takehara, M Yanagihara, H Fujimoto, H Masato, K Inoue ; IEEE Elec Letters; 4/1994 JA Lester, P Huang, M Ahmadi, M Dufault, D Ingram, TH Chen, D Garske, PD Chow ; IEEE MTT-S; 6/1994 M Schlechtweg, W Reinert, A Bangert, J Braunstein, PJ Tasker, R Bosch, WH Haydl, W Bronner, A Huelsmann, K Koehler, J Seibel, R Yu, M Rodwell ; IEEE MTT-S; 6/1994 SW Duncan, A Eskandarian, D Gill, B Golja, B Power, DW Tu, S Svensson, S Weinreb, M Zimmerman, N Byer ; IEEE MTT-S; 6/1994 M Faulkner, P Chye, R Hansen ; IEEE MTT-S; 6/1994 H Okazaki, K Horikawa, M Tanaka ; IEEE MTT-S; 6/1994 K Tateoka, A Sugimura, H Furukawa, M Yuri, N Yoshikawa, K Kanazawa ; IEEE MTT-S; 6/1994 L Marosi, D Yamauchi, J Goel, G Onak, DI Stones, A Sharma, K Tan, B Brunner, J Mancini ; IEEE MTTS; 6/1994 M Rofougaran, A Rofougaran, C Olgaard, AA Abidi ; IEEE VLSI; 6/1994 FM Ghannouchi, GX Zhao, F Beauregard ; IEEE ; 6/1994 122 [245] [246] [247] [248] [249] [250] [251] [252] [253] [254] [255] [256] [257] [258] [259] [260] [261] [262] [263] [264] [265] [266] [267] [268] [269] [270] [271] [272] [273] [274] [275] [276] [277] [278] [279] [280] [281] [282] [283] [284] [285] [286] [287] [288] [289] [290] [291] [292] [293] [294] P Berini, M Desgagne, FM Ghannouchi, RG Bosisio ; IEEE ; 6/1994 RM Wohlert, G Jackson, MG Adlerstein ; IEEE ; 6/1994 M Campovecchio, B Le Bras, M Lajugie, J Obregon ; IEEE ; 6/1994 A Khatibzadeh, W Liu, T Henderson, J Sweder, S Pierce ; IEEE ; 6/1994 F Ali, A Gupta, M Salib, B Veasel, D Dawson ; IEEE ; 6/1994 AK Sharma, cG Onak, R Lai, KL Tan ; IEEE ; 6/1994 K Sakuno, M Akagi, H Sato, M Miyauchi, M Hasegawa, T Yoshimasu, S Hara ; IEEE; 6/1994 T Yoshimasu, N Tanba, S Hara ; IEEE ; 6/1994 T Kunihisa, T Yokoyama, H Fujimoto, K Ishida, H Takehara, O Ishikawa ; IEEE; 6/1994 S Makioka, K Tateoka, M Yuri, N Yoshikawa, K Kanazawa; IEEE; 6/1994 M Ida, H Suzaki, T Tamanaka, T Nishikawa ; IEEE ; 6/1994 AK Sharma, G Onak, D Yamauchi, DI Stones, J Goel, R Lai, KL Tan ; IEEE ; 6/1994 PM Smith, CT Creamer, WF Kopp, DW Ferguson, P Ho, JR Willhite ; IEEE ; 6/1994 TH Chen, PD Chow, KL Tan, JA Lester, G Zell, M Huang ; IEEE ; 6/1994 B Kraemer, R Basset, C Baughman, P Chye, D Day, J Wei; IEEE; 6/1994 K Ishii, T Okamoto, H Ishida, N Tanibe, Y Dooi, H Maeda, M Shigaki, T Katoh ; IEEE ; 6/1994 S Toyoda ; IEEE ; 6/1994 J Shu, J Wei, R Basset, Y Chung, C Hua, C Meng, P Chye, J Hall, D Day; IEEE; 6/1994 JS Cardinal, FM Ghannouchi ; IEEE ; 6/1994 M Nagaoka, T Inoue, K Kawakyu, S Obayashi, H Kayano, E Takagi, Y Tanabe, M Yoshimura, K Ishida, Y Kitaura, N Uchitomi ; IEEE ; 6/1994 W Boulais, RS Donahue, A Platzker, J Huang, L Aucoin, S Shanfield, M Vafidaes ; IEEE ; 6/1994 S Bouthillette, A Platzker, L Aucoin ; IEEE ; 6/1994 B Kraemer, R Basset, P Chye, D Day, J Wei; IEEE; 6/1994 P Chinoy, N Jain, P Li, J Goodrich, C Souchuns ; IEEE ; 6/1994 A Werthof, T Grave, W Kellner ; IEEE MTT-S; 6/1994 SC Wang, MY Kao, SMJ Liu, P Ho, KG Duh ; IEEE DRC; 6/1994 JO Plouchart, H Wang, C Pinatel, M Riet, P Berdaguer, C Doubon-Chevallier ; IEEE MTT-S; 6/1994 TH Wang, CC Liang, JS Hsu, CS Wu ; IEEE EDMS; 7/1994 W Lam, M Matloubian, A Igawa, C Chou, A Kurdoghlian, C Ngo, L Jelloian, A Brown, M Thompson, L Larson; IEEE; 7/1994 W Liu, A Khatibzadeh, T Kim, J Sweder ; IEEE; 9/1994 RE Kasody, GS Dow, AK Sharma, MV Aust, D Yamauchi, R Lai, M Biedenbender, KL Tan, BR Allen; IEEE; 9/1994 M Maeda, M Nishijima, H Takehara, C Adachi, H Fujimoto, 0 Ishikawa ; IEEE J SS Circ; 10/1994 NL Wang, WJ Ho, JA Higgins ; IEEE; 10/1994 CE Weitzel, JW Palmour, CH Carter, KJ Nordquist; IEEE EDL; 10/1994 JJ Komiak ; IEEE Workshop Nonlin MW; 10/1994 M Campovecchio, R Hilal, B Le Bras, M Lajugie, J Obregon ; IEEE Workshop Nonlin MW; 10/1994 C Duvanaud, P Bouysse, S Dietsche, JM Nebus, JM Paillot, D Roques ; IEEE Workshop Nonlin MW; 10/1994 C Pinatel, S Vuye, C Dubon-Chevalier, H Wang ; IEEE Workshop Nonlin MW; 10/1994 CE Weitzel, JW Palmour, CH Carter, KJ Nordquist ; IEEE EDL; 10/1994 T Sowlati, CAT Salama, J Sitch, G Rabjohn, D Smith; IEEE GaAs IC; 10/1994 J Naber, J Griffiths, J Salvey ; IEEE GaAs IC; 10/1994 T Quach, J Staudinger ; IEEE GaAs IC; 10/1994 JJ Komiak, LW Yang ; IEEE GaAs IC; 10/1994 H Wang, GS Dow, M Aust, KW Chang, R Lai, M Biedenbender, DC Streit, BR Allen; IEEE GaAs IC; 10/1994 LW Yang, JJ Komiak, MY Kao, DE Houston, DP Smith, KJ Norheden; IEEE IEDM; 12/1994 AK Sharma, GP Onak, R Lai, KL Tan ; IEEE; 12/1994 F Ali, A Gupta, M Salib, BW Veasel, DE Dawson ; IEEE Trans MTT; 12/1994 K Matsunaga, Y Okamoto, M Kuzuhara; IEEE IEDM; 12/1994 Y Tateno, H Yamada, S Ohara, S Kato, H ohnishi, T Fujii, J Fukaya ; IEEE IEDM; 12/1994 T Yokoyama, T Kunihisa, H Fujimoto, H Takehara, K Ishida, H Ikeda, O Ishikawa; IEEE Trans MTT; 12/1994 123 [295] [296] [297] [298] [299] [300] [301] [302] [303] [304] [305] [306] [307] [308] [309] [310] [311] [312] [313] [314] [315] [316] [317] [318] [319] [320] [321] [322] [323] [324] [325] [326] [327] [328] [329] [330] [331] [332] [333] [334] [335] [336] [337] [338] [339] [340] M Aust, H Wang, M Biedenbender, R Lai, DC Streit, PH Liu, GS Dow, BR Allen ; IEEE MWGW Letters; 1/1995 JCL Chi, JA Lester, Y Hwang, PD Chow, MY Huang ; IEEE MWGW Letters; 1/1995 S Tanaka, E Hase, A Nakajima, K Sugano, T Fujioka, Y Imakado, K Fujiwara, T Okamoto, Y Shigeno, K Sato, I Arai, M Yamane, C Kusano, K Sakamoto, J Nakagawa, M Koya; IEEE ISSCC; 1/1995 CS Wu, CK pao, W Yau, H Kanber, M Hu, SX Bar, A Kurdoghlian, Z Bardai, D Bosch, C Seashore, M Gawronski ; IEEE Trans MTT; 1/1995 FM Ghannouchi, F Beauregard, GX Zhao ; IEEE IMTC; 4/1995 DC Miller, RA Sadler, AH Peake ; IEEE MTT-S; 6/1995 H Wang, Y Hwang, TH Chen, M Biedenbender, DC Streit, DCW Lo, GS Dow, BR Allen ; IEEE MTT-S; 6/1995 J Hubert, J Schoenberg, ZB popovic ; IEEE MTT-S; 6/1995 L Marosi, M Sholley, J Goel, A Faris, M Siddiqui, DI Stones, K Tan ; IEEE MTT-S; 6/1995 M Kumar, M Hanczor, H Voigt, G Cambigianis, R Sachs, C Bonilla; IEEE MTT-S; 6/1995 M Knox, R Youmans, M Kumar, L Forker, H Voigt, P Brand; IEEE MTT-S; 6/1995 Y Kalayci, R Tempel, W Luetke, M Akpinar, I Wolff; IEEE MTT-S; 6/1995 S Weinreb, E Fischer, B Kane, N Byer, M Zimmerman ; IEEE MMWMCS; 6/1995 BT Agar, DS Bowser, KV Buer, DW Corman, CD Grondahl ; IEEE MMWMCS; 6/1995 JD Huang, D Zhang ; IEEE MTT-S; 6/1995 JM Dieudonne, B Adelseck, P Narozny, H Daembkes ; IEEE MTT-S; 6/1995 SW Chen, PM Smith, SMJ Liu, WF Kopp, TJ Rogers ; IEEE MWGW Letters; 6/1995 J Schoenberg, T Mader, B Shaw, ZB Popovic ; IEEE MTT-S; 6/1995 G Jackson, D Teeter, D Bradford, M Cobb ; IEEE MTT-S; 6/1995 A Platzker, S Bouthillette ; IEEE MTT-S; 6/1995 K Matsunaga, Y Okamoto, M Kuzuhara ; IEEE MTT-S; 6/1995 G Ferrell, L Dickens, J Gipprich, B Hayes, F Sacks ; IEEE MTT-S; 6/1995 F Ali, A Gupta, M Salib, B Veasel ; IEEE MTT-S; 6/1995 S Toyada ; IEEE MTT-S; 6/1995 M Maeda, H Takehara, M Nakamura, Y Ota, 0 Ishikawa ; IEEE MTT-S; 6/1995 SJ Maeng, CS Lee, KH Youn, JL Lee, HM Park ; IEEE MTT-S; 6/1995 H Masato, M Maeda, H Fujimoto, S Morimoto, M Nakamura, Y Yoshikawa, H Ikeda, H Kosugi, Y Ota; IEEE MTT-S; 6/1995 J Griffiths, V Sadhir ; IEEE MTT-S; 6/1995 M Muraguchi, M Nakatsugawa, H Hayashi, M Aikawa ; IEEE MTT-S; 6/1995 JJ Komiak, LW Yang ; IEEE MTT-S; 6/1995 S Sriram, R Barron, AW Morse, TJ Smith, G Augustine, AA Burk, RC Clarke, RC Glass, HM Hobgood, PA Orphanos, RR Siergiej, CD Brandt, MC Driver, RH Hopkins ; IEEE DRC; 6/1995 N Yoshikawa, K Tateoka, K Miyatsuji, S Makioka, K Kanazawa; IEEE ISSCC; 6/1995 RG Freitag, HG Henry, EK Lee, M Pingor, HL Salvo ; IEEE Trans MTT; 7/1995 JS Cardinal, FM Ghannouchi ; IEEE Trans MTT; 7/1995 PM Smith, SMJ Liu, MY Kao, P Ho, SC Wang, KHG Duh, ST Fu, PC Chao ; IEEE MWWC Letters; 7/1995 RC Clarke, RR Siergiej, AK Agarwal, CD Brandt, AA Burk, A Morse, PA Orphanos ; IEEE HSSDC; 7/1995 D Lovelace, D Ngo, J Costa, N Camilleri ; IEEE PIMRC; 9/1995 T Mader, M Markovic, ZB Popovic ; IEEE PIMRC; 9/1995 T Sowlati, CAT Salama, J Sitch, G Rabjohn, D Smith ; IEEE J SS Circuits; 10/1995 H Wang, R Lai, M Biedenbender, GS Dow, BR Allen ; IEEE J SS Circuits; 10/1995 Y Itoh ; IEEE ISSSE; 10/1995 K Ohata, T Saito; IEEE ISSSE; 10/1995 A Martin, A Mortazawi, BC De Loach ; IEEE MWGW Letters; 10/1995 Y Hasegawa, Y Ogata, I Nagasaki, K Inosaki, N Iwata, M Kanamori, T Itoh ; IEEE GaAs IC; 10/1995 P Narozny, H Tobler, GK Kornfeld, RA Nunn, B Adelseck, M Ludwig, G Eggers; IEEE GaAs IC; 10/1995 H Hayashi, M Nakatsugawa, M Muraguchi ; IEEE Trans MTT; 10/1995 124 [341] K Kamozaki, S Tanaka, E Hase, A Nakajima, T Fujioka, K Sugano, Y Imakado, K Fujiwara, T Okamoto, K Sato, Y Shigeno, I Arai, M Yamane, H Kundoh ; IEEE Conf on Universal Personal Communications; [342] [343] [344] N Kasai, M Noda, K Ito, K Yamamoto, K Maemura, Y Ohta, T Ishikawa, Y Yoshii, M Nakayama, H Takano, O Ishihara; IEEE GaAs IC; 10/1995 P Walters, P Lau, K Buehring, J Penney, C Farley, P McDade, K Weller ; IEEE GaAs IC; 10/1995 R Lai, M Biedenbender, J Lee, K Tan, D Streit, PH Liu, M Hoppe, B Allen ; IEEE GaAs IC; 10/1995 [345] PM White, TM O'Leary; IEEE GaAs IC; 10/1995 [346] [347] [348] [349] [350] YC Chen, CS Wu, CK Pao, M Cole, Z Bardai, LD Hou, TA Midford, TC Cisco ; IEEE GaAs IC; 10/1995 JM Schellenberg ; IEEE GaAs IC; 10/1995 H Wang, M Sayed ; IEEE MWGW Letters; 12/1995 I Ohbu, T Tanimoto, S Tanaka, H Matsumoto, A Terano, M Kudo, T Nakamura ; IEEE IEDM; 12/1995 RR Siergiej, RC Clarke, AK Agarwal, CD Brandt, AA Burk, A Morse, PA Orphanos ; IEEE IEDM; 12/1995 S Ohara, H Yamada, T Iwai, Y Yamaguchi, K Imanishi, K Joshin ; IEEE IEDM; 12/1995 K Riepe, H Leier, U Seiler, A Marten, H Sledzik ; IEEE IEDM; 12/1995 N Constantin, FM Ghannouchi ; IEEE Trans MTT; 12/1995 M Maeda, H Masato, H Takehara, M Nakamura, S Morimoto, H Fujimoto, Y Ota, O Ishikawa ; IEEE Trans MTT; 12/1995 U Erben, M Wahl, A Schueppen, H Schumacher ; IEEE MWGW Letters; 12/1995 Y Itoh, M mochizuki, M Kohno, H Masuno, T Tagaki, Y Mitsui ; IEEE MWGW Letters; 12/1995 SK Wang, CD Chang, M Siracusa, LCT Liu, RG Pauley, P Asher, M Sokolich ; IEEE Trans MTT; 12/1995 A Gupta, F Ali, D Dawson, P Smith ; IEEE MWGW Letters; 1/1996 MJ McCullagh, NB Roberts ; IEEE SSPAG; 1/1996 K Riepe, H Leier, U Seiler, A Marten, H Sledzik ; IEEE MWGW Letters; 1/1996 F Deshours, E Bergeault, G Berghoff, C Pinatel, C Dubon-Chevallier ; IEEE MWGW Letters; 1/1996 K Yamamoto, K Maemura, Y Ohta, N Kasai, M Noda, H Yuura, Y Yoshii, M Nakayama, N Ogata, T Takagi, M Otsubo; IEEE ISSCC; 1/1996 JJ Brown, JA Pusl, M Hu, AE Schmitz, DP Docter, JB Shealy, MG Case, MA Thompson, LD Nguyen; IEEE MWGW Letters; 1/1996 PC Abernathy, JR Dobrazanski, MC Smith, MS Griswold, MG Brunsed, JR Sherman ; IEEE Southeastcon; 4/1996 S Makioka, N Yoshikawa, K Kanazawa; IEEE Trans MTT; 4/1996 KM Simon, RM Wohlert, JP Wendler, LM Aucoin, DW Vye ; IEEE MMWMCS; 6/1996 M Nakayama, K Mori, N Ogata, Y Mitsui, H Yuura, Y Yoshii, K Yamamoto, K Maemura, O Ishida ; IEEE MTT-S; 6/1996 W Titus, M Shifrin, V Aparin, B Bedard; IEEE MMWMCS; 6/1996 JA Lester, Y Hwang, J Chi, R Lai, M Biedenbender, PD Chow ; IEEE MMWMCS; 6/1996 J Goel, G Onak, DI Stones, D Yamauchi, A Sharma, K Tan, J Mancini ; IEEE MTT-S; 6/1996 S Bouthillette, A Platzker ; IEEE MTT-S; 6/1996 S Makioka, S enomoto, H Furukawa, K Tateoka, N Yoshikawa, K Kanazawa; IEEE MTT-S; 6/1996 V Jair, S Tehrani, D Halchin, E Glass, E Fisk, M Majerus ; IEEE MTT-S; 6/1996 R Yarborough, P Saunier, HQ Tserng ; IEEE MTT-S; 6/1996 WL Pribble, EL Griffin ; IEEE MTT-S; 6/1996 M Cardullo, C Page, D Teeter, A Platzker ; IEEE MTT-S; 6/1996 Y Hwang, PI) Chow, J Lester, J Chi, D Garske, M Biedenbender, R Lai ; IEEE MTT-S; 6/1996 B Nelson, S Cripps, J Stevenson Kenney, AF Podell ; IEEE MTT-S; 6/1996 J Sweder, G .Austin Truitt, SR Nelson, JS Mason ; IEEE MTT-S; 6/1996 H Ono, Y Umemoto, M Mori, M Miyazaki, A Terano, M Kudo ; IEEE MTT-S; 6/1996 H Yamada, S Ohara, T Iwai, Y yamaguchi, K Imanishi, K Joshin ; IEEE MTT-S; 6/1996 S Toyoda ; IEEE MTT-S; 6/1996 JA Pusl, JJ Brown, JB Shealy, M Hu, AE Schmitz, DP Docter, MG Case, MA Thompson, LD Nguyen; IEEE MTT-S; 6/1996 K Matsunaga, Y Okamoto, I Miura, M Kuzuhara ; IEEE MTT-S; 6/1996 MK Siddiqui, AK Sharma, LG Callejo, CH Chen, K Tan, HC Yen ; IEEE MTT-S; 6/1996 DW Wu ; IEEE MTT-S; 6/1996 10/1995 [351] [352] [353] [354] [355] [356] [357] [358] [359] [360] [361] [362] [363] [364] [365] [366] [367] [368] [369] [370] [371] [372] [373] [374] [375] [376] [377] [378] [379] [380] [381] [382] [383] [384] [385] [386] 125 [387] [388] [389] [390] [391] [392] [393] [394] [395] [396] [397] [398] [399] [400] [401] [402] [403] [404] [405] [406] [407] [408] [409] [410] [411] [412] [413] [414] [415] [416] [417] [418] [419] [420] [421] [422] [423] [424] [425] S Tanaka, S Murakami, Y Amamiya, H Shimawaki, N Furuhata, N Goto, K Honjo, Y Ishida, Y Saito, K Yamamoto, M Yajima, R Temino, Y Hisada ; IEEE MTT-S; 6/1996 A Mallet, T Peyretaillade, R Sommet, D Floriot, S Delage, JM Nebus, J Obregon ; IEEE MTT-S; 6/1996 B Ingruber, W Pritzl, G Magerl ; IEEE MTT-S; 6/1996 T Sowlati, Y Greshishchev, C Andre, T Salama ; IEEE Elec Letters; 9/1996 TM Roh, Y Suh, B Kim, W Park, JB Lee, YS Kim, GY Lee ; IEEE Elec Letters; 9/1996 JB Vincent, DG van der Merwe ; IEEE AFRICON; 9/1996 R Lai, GI Ng, DCW Lo, T Block, H Wang, M Biedenbender, D Streit, PH Liu, RM Dia, EW Lin, HC Yen; IEEE MWGW Letters; 10/1996 L Sjogren, J Choe, D Garske, R Duprey, K Kono, L Ma, D Tamura, C Gandhi, E Boyd, PD Chow, L Rowe ; IEEE ISPAST; 10/1996 R Wang, M Cole, LD Hou, P Chu, CD Chang, TA Midford, T Cisco ; IEEE GaAs IC; 10/1996 R Lai, M Nishimoto, Y Hwang, M Biedenbender, B Kasody, C Geiger, YC Chen, G Zell; IEEE GaAs IC; 10/1996 RJ Trew, MW Shin, V Gatto ; IEEE GaAs IC; 10/1996 S Murakami, S Tanaka, Y Amamiya, H Shimawaki, N Goto, K Honjo, Y Ishida, Y Saitoh, M Yajima, Y Hisada; IEEE GaAs IC; 10/1996 OSA Tang, KHG Duh, SMJ Liu, PM Smith, WF Kopp, TJ Rogers, DJ Pritchard ; IEEE GaAs IC; 10/1996 M Hirose, K Nishihori, M Nagaoka, Y Ikeda, A Kameyama, Y Kitaura, N Uchitomi ; IEEE GaAs IC; 10/1996 MJ Martinez, E Schirmann, M Durlam, D Halchin, R Burton, JH Huang, S Tehrani, A Reyes, D Green, N Cody ; IEEE GaAs IC; 10/1996 MF O'Keefe, GN Henderson, TE Boles, P Noonan, JM Sledziewski, B Brown ; IEEE EDMO; 10/1996 M Yajima, Y Hisada, Y Ishida, Y Saito, K Yamamoto, S Murakami, S Tanaka, N Goto, K Honjo ; IEEE EDMO; 10/1996 K Yamamoto, K Maemura, N Kasai, Y Yoshii, Miyazaki, M Nakayama, N Ogata, T Takagi, M Otsubo; IEEE J SS Circ; 12/1996 A Wood, C Dragon, W Burger ; IEEE IEDM; 12/1996 AK Agarwal, G Augustine, V Balakrishna, CD Brandt, AA Burk, LS Chen, RC Clarke, PM Esker, HM Hobgood, RH Hopkins, AW Morse, LB Rowland, S Seshadri, RR Siergiej, TJ Smith, S Sriram; IEEE IEDM; 12/1996 L Andrieux, A Cazarre, JP Bailbe, A Marty ; IEEE Proc CDS; 12/1996 H Yamada, S Ohara, T Iwai, Y Yamaguchi, K Imanishi, K Joshin ; IEEE Trans MTT; 12/1996 E Schirmann, R Baeten, J Costa, DL Green, D Halchin, MJ Martinez, A Reyes ; IEEE MTTWAD; 1/1997 S Yamamoto, T Yokoyama, T Kunihisa, M Nishijima, M Nishitsuji, K Nishii, O Ishikawa; IEEE MTTWAD; 1/1997 BJ Bulk, MW Green, AW Dearn, AP Long, SP Melvin, LM Devlin, JC Clifton ; IEEE RFMCCWA; 1/1997 D Hill, R Yarborough, T Kim, HF Chau ; IEEE MWGW Letters; 1/1997 V Radisic, ST Chew, Y Qian, T Itoh ; IEEE MWGW Letters; 1/1997 HP Trost, W Simbuerger, DH Wohlmuth, H Knapp, P Weger, AL Scholz ; IEEE RF and MW Circ for Comm WL Appl; 1/1997 T Tanaka, H Furukawa, H Takenaka, T Ueda, T Fukui, D Ueda ; IEEE Trans MTT; 3/1997 CW Kim, N Hayama, N Goto, K Honjo ; IEEE EDL; 4/1997 T Sowlati, YM Greshishchev, CAT Salama ; IEEE J of Solid State Circ; 4/1997 YC Chen, R Lai, E Lin, H Wang, T Block, HC Yen, D Streit, W Jones, PH Liu, RM Dia, TW Huang, PP Huang, K Stamper ; IEEE MWGW Letters; 4/1997 YC Chen, R Lai, H Wang, HC Yen, D Streit, RM Dia, W Jones, T Block, PH Liu, TW Huang, YC Chou, K Stamper ; IEEE IPRM; 4/1997 WL Chen, TS Kim, HF Chau, T Henderson ; IEEE IPRM; 4/1997 D Su, W McFarland ; IEEE Custom IC; 4/1997 Y Hwang, J lester, G Schreyer, G Zell, S Schrier, D Yamauchi, G Onak, B Kasody, R Kono, YC CHen, R Lai ; IEEE MTT-S; 6/1997 MR HeDann, M Jones, G Wilcox, J McLeod, SC Miller ; IEEE MTT-S; 6/1997 AR Barnes, MT Moore, MB Allenson ; IEEE MTT-S; 6/1997 MD Pollman, C Trantanella, M Shifrin, V Aparin, D Upton; IEEE MTT-S; 6/1997 126 [426] [427] [428] [429] [430] [431] [432] [433] [434] [435] [436] [437] [438] [439] [440] [441] [442] [443] [444] [445] [446] [447] [448] [449] [450] [451] [452] [453] [454] [455] [456] [457] [458] [459] [460] [461] [462] [463] [464] [465] [466] JA Lester, J Chi, R Lai, M Biedenbender, D Garske, R Rordan, PD Chow ; IEEE MTT-S; 6/1997 H Kayano, S Obayashi, T Maeda, Y Suzuki; IEEE MTT-S; 6/1997 YC Wang, JM Kuo, JR Lothian, F Ren, HS Tsai, JS Weiner, J Lin, A Tate, WE Mayo, YK Chen ; IEEE DRC; 6/1997 H Izadpanah, RC Malkemes, DD Chukurov, RR Cordell ; IEEE Trans CircSys; 6/1997 P Huang, E Lin, R Lai, M Biedenbender, TW Huang, H Wang, C Geiger, T Block, PH Liu ; IEEE MTT-S; 6/1997 DL Ingram, DI Stones, TW Huang, M Nishimoto, H Wang, M Siddiqui, D Tamura, J Elliott, R Lai, M Biedenbender, HC Yen, B Allen; IEEE MTT-S; 6/1997 GN Henderson, MF O'Keefe, TE Boles, P Noonan, JM Sledziewski, BM Brown ; IEEE MTT-S; 6/1997 M Kimishima, K Hayashi, M Takahashi; IEEE MTT-S; 6/1997 W Abey, T Kawai, I Okamoto, M Suzuki, C Khandavalli, W Kennan, Y Tateno, M Nagahara, M Takikawa ;IEEE MTT-S; 6/1997 DW Wu, R Parkhurst, SL Fu, J Wei, CY Su, SS Chang, D Moy, W Fields, P Chye, R Levitsky ; IEEE MTT-S; 6/1997 M Nagaoka, H Wakimoto, K Kawakyu, K Nishihori, Y Kitaura, T Sasaki, A Kameyama, N Uchitomi; IEEE MTT-S; 6/1997 E Glass, JH Huang, J Abrokwah, B Bernhardt, M Majerus, E Spears, R Droopad, B Ooms ; IEEE MTT-S; 6/1997 K Yamaguchi, TB Nishimura, N Iwata, T Takemura, M Kuzuhara, Y Miyasaka ; IEEE MTT-S; 6/1997 J Morikawa, K Asano, K Ishikura, H Oikawa, M Kanamori, M Kuzuhara ; IEEE MTT-S; 6/1997 I Takenaka, H Takahashi, K Asano, J Morikawa, K Ishikura, M Kanamori, M Kuzuhara, H Tsutsui ; IEEE MTT-S; 6/1997 JJ Komiak, SC Wang, TJ Rogers ; IEEE MTT-S; 6/1997 S Toyoda ; IEEE MTT-S; 6/1997 V Radisic, Y Qian, T Itoh ; IEEE MTT-S; 6/1997 KI Jeon, YS Kwon, SC Hong ; IEEE MTT-S; 6/1997 P Cameron, W Pan, C hanz, R Nicklaus, P Chu, D Wong, T Cisco; IEEE MTT-S; 6/1997 JP Viaud, M Lajugie, R Quere, J Obregon ; IEEE MTT-S; 6/1997 RJ Trew ; IEEE MTT-S; 6/1997 S Sriram, TJ Smith, LB Rowland, AA Burk, G Augustine, V Balakrishna, HM Hobgood, CD Brandt; IEEE DRC; 6/1997 S Tanaka, Y Amamiya, S Murakami, H Shimawaki, N Goto, Y Takayama, K Honjo; IEEE MMWaves; 7/1997 T Inoue, K Hosoya, K Ohata ; IEEE MMWaves; 7/1997 B Bayraktaroglu, M Salib ; IEEE MWGW Letters; 7/1997 Y Tkachenko, D Bartle, P DiCarlo, D Mitchell, D Petzold ; IEEE WCC; 7/1997 YL Lai, EY Chang, CY Chang, TH Liu, SP Wang, HT Hsu ; IEEE MWGW Letters; 7/1997 FM Rotella, Z Yu, R Dutton, B Troyanovsky, G Ma ; IEEE SISPAD; 9/1997 YL Lai, EY Chang, CY Chang, MC Tai, TH Liu, SP Wang, KC Chuang, CT Lee ; IEEE EDL; 9/1997 HF Chau, G Wilcox, W Chen, M Tutt, T Henderson ; IEEE MWGW Letters; 9/1997 OSA Tang, KHG Duh, SMJ Liu, PM Smith, WF Kopp, TJ Rogers, DJ Pritchard ; IEEE J SS Circ; 9/1997 JE Mueller, T Grave, HJ Siweris, M Kaerner, A Schaefer, H Tischer, H Riechert, L Schleicher, L Verweyen, A Bangert, W Kellner, T Meier; IEEE J SS Circ; 9/1997 Ph Eudeline, P Perez, JP Lemette ; IEEE Radar; 10/1997 YA Tkachenko, CJ Wei, D Bartle ; IEEE GaAs Reliab Wksp; 10/1997 RS Virk, MY Chen, C Nguyen, T Liu, M Matloubian, DB Rensch ; IEEE MWGW Letters; 10/1997 M Nagoka, H Wakimoto, T Seshita, K Kawakyu, Y Kitaura, A Kameyama, N Uchitomi ; IEEE GaAs IC; 10/1997 JE Mueller, P Baureis, O Berger, T Boettner, N Bovoloon, G Packeiser, P Zwicknagl ; IEEE GaAs IC; 10/1997 K Kamozaki, N Kurita, W Hioe, T Tanimoto, H Ohta, T Nakamura, H Kondoh ; IEEE GaAs IC; 10/1997 T Yoshimasu, M Akagi, N Tanba, S Hara ; IEEE GaAs IC; 10/1997 DR Greenberg, M Rivier, P Girard, E Bergeault, J Moniz, D Ahlgren, G Freeman, S Subbanna, SJ Jeng, K Stein, D Nguyen-Ngoc, K Schonenberg, J Malinowski, D Colavito, DL Harame, B Meyerson ; IEEE IEDM; 12/1997 127 Y Itoh, A Tsuchiko, M Kasahara, K Inami, J Udomoto, S Tsuji ; IEEE APMC; 12/1997 JW Lee, WY Kim, YS Kim, TM Rho, BM Kim ; IEEE Trans MTT; 12/1997 D Hill, TS Kim ; IEEE Trans MTT; 12/1997 PP Huang, TW Huang, H Wang, EW Lin, YH Shu, GS Dow, R Lai, M Biedenbender, JH Elliott ; IEEE Trans MTT; 12/1997 [471] DL Ingram, DI Stones, JH Elliott, H Wang, R Lai, M Biedenbender ; IEEE Trans MTT; 12/1997 [472] I Yoshida ; IEEE IEDM; 12/1997 [473] Q Ren, Y Gao, I Wolff ; IEEE Asia Pacific MW Conf; 12/1997 [474] S Piotrowicz, C Gaquiere, B Bonte, E Bourcier, D Theron, X Wallart, Y Crosnier ; IEEE MWGW Letters; 1/1998 [475] V Radisic, Y Qian, T Itoh ; IEEE MWGW Letters; 1/1998 [476] S Tanaka, Y Amamiya, S Murakami, H Shimawaki, N Goto, Y Takayama, K Honjo ; IEEE TED; 1/1998 [477] J Portilla, H Garcia, E Artal ; IEEE J SS Circ; 1/1998 [478] K Yamamoto, T Moriwaki, Y Yoshii, T Fujii, J Otsuji, Y Sasaki, Y Miyazaki, K Nishitani ; IEEE ASPDAC; 1/1998 [479] I Yoshida, M Katsueda, M Morikawa, Y Matsunaga, T Fujioka, M Hotta, Y Nunogawa, K Kobayashi, S Shimuzu, M Nagata ; IEEE ISSCC; 1/1998 [480] RC Clarke, CD Brandt, S Sriram, RR Siergiej, AW Morse, AK Agarwal, LS Chen, V Balakrishna, AA Burk; IEEE DSC; 1/1998 [481] Q Chen, JW Yang, R Gaska, MA Khan, MS Shur, GJ Sullivan, AL Sailor, JA Higgins, AT Ping, I Adesida ; IEEE EDL; 1/1998 [482] K Yamaguchi, N Iwata, M Kuzuhara, T Takayame ; IEEE TED; 1/1998 [483] JG McRory, GG Rabjohn, RH Johnston ; IEEE J SS Circ; 1/1998 [484] A Rofougaran, G Chang, JJ Rael, JYC Chang, M Rofougaran, PJ Chang, M Djafari, MK Ku, EW Roth, AA Abidi, H Samueli ; IEEE J SS Circ; 4/1998 [485] JN Burghartz, JO Plouchart, KA Jenkins, CS Webster, M Soyuer ; IEEE EDL; 4/1998 [486] W Wojtasiak, D Gryglewski, T Morawski, E Sedek; IEEE ICMR; 4/1998 [487] R Goetzfried, F Beisswanger, S Gerlach, A Schueppen, H Dietrich, U Seiler, KH Bach, J Albers ; IEEE Trans MTT; 4/1998 [488] SS Taylor; IEEE MTT-S; 6/1998 [489] HN Ma, R Singh, KT Yan, SJ Fang, FJ Lin, KS Tan, J Shibata, A Tamura, H Nakamura ; IEEE MTT-S; 6/1998 [490] M Nakayama, K Horiguchi, K Yamamoto, Y Yoshii, S Sugiyama, N Suematsu, T Takagi ; IEEE MTT-S; 6/1998 [491] HJ Siweris, A Werthof, H Tischer, U Schaper, A Schaefer, L Verweyen, T Grave, G Bloeck, M Schlechtweg, W Kellner ; IEEE MTT-S; 6/1998 [492] T Yokoyama, M Nishijima, T Kunihisa, S Yamamoto, 0 Ishikawa ; IEEE MTT-S; 6/1998 [493] M Komaru, H Hoshi, H Kurusu, Y Notani, T Katoh, T Ishida, T Oku, T Ishikawa, Y Mitsui ; IEEE MTT-S; 6/1998 [494] SHL Tu, C Toumazou ; IEEE ISCAS; 6/1998 [495] J Shu, T Hwang, D Nguyen, R Pumares, P Chye, P Khanna ; IEEE MTT-S; 6/1998 [496] A Wood, W brakensiek, C Dragon, W Burger ; IEEE MTT-S; 6/1998 [497] H Ashoka, J Ness, A Robinson, M Gourlay, J Logan, P Woodhead, D reuther ; IEEE MTT-S; 6/1998 [498] S Wong, S Luo, L Hadley ; IEEE ISSCC; 6/1998 [499] T Tanimoto, I Ohbu, S Tanaka, A Kawai, M Kudo, A Terano, T Nakamura ; IEEE TED; 6/1998 [500] T Iwai, S Ohara, H Yamada, Y Yamaguchi, K Imanishi, K Joshin ; IEEE TED; 6/1998 [501] T Budka, A Ketterson, HQ Tserng, L Stiborek, L Heinrich, R Smith, B Ables, C Kyhl, K Worthen, G Brehm, J Reddick ; IEEE MWGW Letters; 6/1998 [502] AL Martin, A Mortazawi, BC DeLoach ; IEEE Trans MTT; 6/1998 [503] B Ingruber, W Pritzl, D Smely, M Wachutka, G Magerl ; IEEE Trans MTT; 6/1998 [504] T Seshita, K Kawakyu, H Wakimoto, M Nagaoka, Y Kitaura, N Uchitomi ; IEEE MTT-S; 6/1998 [505] H Asano, S Hara, S Komai ; IEEE MTT-S; 6/1998 [506] DL Ingram, L Sjogren, J Kraus, M Nishimoto, M Siddiqui, S Sing, K Cha, M Huang, R Lau ; IEEE MTTS; 6/1998 [507] DC Yang, JM Yang, H Wang, P Huang ; IEEE MTT-S; 6/1998 [508] D Smely, B Ingruber, M Wachutka, G Magerl ; IEEE MTT-S; 6/1998 [467] [468] [469] [470] 128 [509] [510] [511] [512] [513] [514] [515] [516] [517] [518] [519] [520] [521] [522] [523] [524] [525] [526] [527] [528] [529] [530] [531] [532] [533] [534] [535] [536] [537] [538] [539] [540] [541] [542] [543] [544] [545] [546] [547] [548] [549] [550] [551] [552] [553] V Radisic, Y Qian, T Itoh ; IEEE MTT-S; 6/1998 T Iwai, S Ohara, T Miyashita, K Joshin ; IEEE MTT-S; 6/1998 M Nishida, S Murai, H Uda, H Tominaga, T Sawai, A Ibaraki ; IEEE MTT-S; 6/1998 TB Nishimura, N Iwata, K Yamaguchi, K Takemura, Y Miyasaka ; IEEE MTT-S; 6/1998 MK Siddiqui, AK Sharma, LG Callejo, R Lai ; IEEE MTT-S; 6/1998 J Schnellenberg ; IEEE MTT-S; 6/1998 M Salib, A Gupta, A Ezis, M Lee, M Murphy ; IEEE MTT-S; 6/1998 JA Pusl, RF Widman, JJ Brown, M Hu, N Kaur, M BeZaire, LD Nguyen ; IEEE MTT-S; 6/1998 H Tsutsui, I Takenaka, H Takahashi, K Asano, J Morikawa, K Ishikura, M Kuzuhara ; IEEE MTT-S; 6/1998 K Nishihori, Y Kitaura, M Hirose, M Mihara, M Nagaoka, N Uchitomi ; IEEE TED; 7/1998 JL Lee, HC Kim, KY Mun, SJ Maeng ; IEEE EDL; 7/1998 P Perugupalli, M Trivedi, K Shenai, SK Leong ; IEEE TED; 7/1998 TA Boes, U Lott, W Baechtold ; IEEE RAWCON; 7/1998 A Wood, W Brakensiek ; IEEE RAWCON; 7/1998 J Staudinger, G Norris, R Sherman, G Sadowniczak ; IEEE RAWCON; 7/1998 E Jaervinen ; IEEE RAWCON; 7/1998 D Helms, M Testa, N Vladimirsky, D Wills ; IEEE RAWCON; 7/1998 YG Kim, SJ Maeng, JH Lee, CS Park ; IEEE RAWCON; 7/1998 IM Abolduyev, AM Zubkov, VM Minnebaev ; IEEE ICSC; 9/1998 R String, A Agarwal, T Smith, R Messham, S Mani, V Hegde, M Hanes, H Nathanson, P Potyraj, K Petrosky, T Knight, P Brabant ; IEEE SMICRF; 9/1998 JE Mueller, P Baureis, O Berger, T Boettner, N Bovolon, R Schultheis, G Packeiser, P Zwicknagel ; IEEE J SS Circ; 9/1998 T Yoshimasu, M Akagi, N Tanba, S Hara ; IEEE J SS Circ; 9/1998 R Goetzfried, F Beisswanger, S Gerlach ; IEEE J SS Circ; 9/1998 TB Mader, EW Bryerton, M Markovic, M Forman, Z Popovic ; IEEE Trans MTT; 10/1998 FJ Ortega-Gonzalez, JL Jimenez-Martin, A Asensio-Lopez, G Torregrosa-Penalva ; IEEE MWGW Letters; 10/1998 RJ Trew ; IEEE ISSEE; 10/1998 D Barataud, C Arnaud, B Thubaud, M Campovecchio, JM Nebus, JP Villotte ; IEEE Trans Inst Meas; 10/1998 M Yu, M Matloubian, P Petre, L Hamilton, R Bowen, M Lui, HC Sun, C Ngo, P Janke, D Baker, R Robertson ; IEEE GaAs IC; 10/1998 JJ Komiak, W Kong, PC Chao, K Nichols; IEEE GaAs IC; 10/1998 S Goto, K Fujii, H Morishige, S Suzuki, S Sakamoto, N Yoshida, N Tanino, K Sato ; IEEE GaAs IC; 10/1998 I Takenaka, H Takahashi, K Asano, K Ishikura, J Morikawa, K Sato, I Takano, K Hasegawa, K Tokunaga, F Emori, M Kuzuhara; IEEE GaAs IC; 10/1998 BY Majlis, A Ariffin, AFA Mat, S Jaafar, S Bujang, MR Yahya ; IEEE ICSE; 10/1998 K Choumei, K Yamamoto, N Kasai, T Moriwaki, Y Yoshii, T Fujii, J Otsuji, Y Miyazaki, N Tanino, K Sato ; IEEE GaAs IC; 10/1998 B Ingruber, J Baumgartner, D Smely, M Wachutka, G Magerl, FA Petz ; IEEE Trans MTT; 10/1998 YC Chen, DL Ingram, R Lai, M Barsky, R Grunbacher, T Block, HC Yen, DC Streit; IEEE MWGW Letters; 10/1998 WR Deal, V Radisic, YX Qian, T Itoh ; IEEE MWGW Letters; 10/1998 GH Yun, HK Park ; IEEE Antennas Propag; 12/1998 DK Su, WJ McFarland; IEEE J SS Circ; 12/1998 SL Wong, S Luo ; IEEE J SS Circ; 12/1998 P Parikh, J Ibbetson, U Mishra, D Docter, M Le, K Kiziloglu, D Grider, J Pusl, D Widman, L Kehias, T Jenkins ; IEEE Trans MTT; 12/1998 MK Siddiqui, AK Sharma, LG Callejo, R Lai ; IEEE Trans MTT; 12/1998 G Berghoff, E Bergeaullt, B Huyart, L Jallet ; IEEE Trans MTT; 12/1998 SS Lu, YW Hsu, CC Meng, LP Chen; IEEE EDL; 1/1999 HS Tsai, P Roux, JL Giguet, YK Chen ; IEEE Tech for Wireless Appl; 1/1999 RG Freitag, KM Renaldo, HG Henry, JE Degenford ; IEEE Tech for Wireless Appl; 1/1999 129 [554] [555] [556] [557] [558] [559] [560] [561] [562] [563] [564] [565] [566] [567] [568] [569] [570] [571] [572] [573] [574] [575] [576] [577] [578] [579] [580] [581] [582] [583] [584] [585] [586] [587] [588] [589] [590] [591] [592] [593] [594] [595] D Streit, A Oki, R Lai, V Medvedev, K Kobayashi ; IEEE Tech for Wireless Appl; 1/1999 M Bopp, M Alles, M Arens, D Eichel, S Gerlach, R Goetzfried, F Gruson, M Kocks, G Krimmer, R Reimann, B Ross, M Siegle, J Zieschang ; IEEE ISSCC; 1/1999 B Ballweber, R Gupta, DJ Allstot; IEEE ISSCC; 1/1999 W Simbuerger, HD Wohlmuth, P Weger; IEEE ISSCC; 1/1999 CF Campbell; IEEE MWGW Letters; 3/1999 LJ Zhang ; IEEE Radar Conf; 4/1999 K Yamamoto, T Moriwaki, T Fujii, J Otsuji, M Miyashita, Y Miyazaki, K Nishitani ; IEEE J SS Circ; 4/1999 ST Sheppard, K Doverspike, WL Pribble, ST Allen, JW Palmour, LT Kehias, TJ Jenkins ; IEEE EDL; 4/1999 YC Wang, M Hong, JM Kuo, JP Mannaerts, HS Tsai, J Kwo, JJ Krajewski, YK Chen, AY Cho ; IEEE Elec Letters; 4/1999 YC Chen, P Chin, D Ingram, R Lai, R Grundbacher, M Barsky, T Block, M Wojtowicz, L Tran, V Medvedev, HC Yen, DC Streit, A Brown ; IEEE IPRM; 4/1999 CH Chen, K Krishnamurthy, S Keller, G parish, M Rodwell, UK Mishra, YF Wu ; IEEE Elec Letters; 4/1999 S Morimoto, M Maeda, T Tokoyama, H Ishida, M Nakamura, Y Ota, D Ueda ; IEEE MTT-S; 6/1999 NS Cheng, TP Dao, MG Case, DB Rensch, RA York ; IEEE MTT-S; 6/1999 K Ishida, H Ikeda, H Kosugi, M Nishijima, T Uwano ; IEEE MTT-S; 6/1999 DW Baker, RS Robertson, RT Kihm, M Matloubian, M Yu, R Bowen ; IEEE MTT-S; 6/1999 JR Lamberg, MJ Gawronski, JJ Geddes, WR Carlyon, RA Hart, GS Dow, EW Holmes, MY Huang ; IEEE MTT-S; 6/1999 J Lulja, H Mattila; IEEE MTT-S; 6/1999 K Nishikawa, K Kamogawa, I Toyoda, T Tokumitsu, M Tanaka; IEEE MTT-S; 6/1999 XW Wang, J Cao, B Liang, ES Khoo, H Nakamura, R Singh; IEEE MTT-S; 6/1999 YF Wu, BJ Thibeault, JJ Xu, RA York, S Keller, BP Keller, UK Mishra ; IEEE DRC; 6/1999 JJ Komiak, W Kong, PC Chao, K Nichols; IEEE MTT-S; 6/1999 MK Siddiqui, AK Sharma LG Callejo, R Lai ; IEEE MTT-S; 6/1999 A Ishimaru, M Maeda, T Yoshida, J Ozaki, S Kamihashi ; IEEE MTT-S; 6/1999 T Satoh, A Betti-Berutto, C Poledrelli, C Khandavalli, J Nikaido, S Kuroda, T Yokoyama, J Fukaya; IEEE MTT-S; 6/1999 J Schellenberg, H Dy-Ky ; IEEE MTT-S; 6/1999 TB Nishimura, N Iwata, G Hau ; IEEE MTT-S; 6/1999 N Miyazawa, H Itoh, Y Nakasha, T Iwai, T Mitashita, S Ohara, K Joshin ; IEEE MTT-S; 6/1999 S Yoshida, Y Wakabayashi, M Kohno, K Uemura; IEEE MTT-S; 6/1999 MD Weiss, Z Popovic ; IEEE MTT-S; 6/1999 CY Hang, V Radisic, YX Qian, T Itoh ; IEEE MTT-S; 6/1999 K Krishnamurthy, SI Long, MJW Rodwell ; IEEE MTT-S; 6/1999 K Yamamoto, T Shimura, T Asada, T Okuda, K Mori, K Choumei, S Suzuki, T Miura, S Fujimoto, R Hattori, H Nakano, K Hosogi, J Otsuji, A Inoue, K Yajima, T Ogata, Y Miyazaki, M Yamanouchi; IEEE MTT-S; 6/1999 T Saso, Y Hasegawa, Y Saito, Y Kakuta; IEEE MTT-S; 6/1999 DL Ingram, YC Chen, J Kraus, B Brunner, B Allen, HC Yen, KF Lau; IEEE MTT-S; 6/1999 J Zhang, HY Jia, PH Tsien, TC Lo ; IEEE EDL; 7/1999 KC Tsai, PR Gray ; IEEE J SS Circ; 7/1999 JJ Xu, YF Wu, S Keller, G Parish, S Heikman, BJ Thubeault, UK Mishra, RA York ; IEEE MWGW Letters; 7/1999 T Melly, AS Porret, CC Enz, M Kayal, E Vittoz ; IEEE LPED; 7/1999 J Staudinger, Gilsdorf, D Newman, G Norris, G Sadowniczak, R Sherman, T Quach, V Wang; IEEE RAWCON; 7/1999 EA Sovero, YW Kwon, DS Deakin, J Hong; IEEE RAWCON; 7/1999 D Barataud, F Blache, A Mallet, PP Bouysse, JM Nebus, JP Villotte, J Obregon, J Verspecht, P Auxemery ; IEEE Trans Inst Meas; 7/1999 YF Wu, RA York, S Keller, BP Keller, UK Mishra ; IEEE MWGW Letters; 7/1999 130 [596] [597] [598] [599] [600] [601] [602] [603] [604] [605] [606] [607] [608] [609] [610] [611] [612] [613] [614] [615] [616] [617] [618] [619] [620] [621] [622] [623] [624] [625] [626] [627] [628] [629] [630] [631] [632] [633] [634] [635] [636] YC Wang, JM Kuo, F Ren, JR Lothian, HS Tsai, JS Weiner, HC Kuo, CH Lin, YK Chen, WE Mayo; IEEE Trans MTT; 7/1999 WR Deal, V Radisic, YX Qian, T Itoh ; IEEE Trans MTT; 7/1999 DA Brown, FM Edwards, PM Jupp, JD Birkbeck, DC Pennington, CR Green, HP Forstner ; IEEE WCNC; 9/1999 PD Tseng, LY Zhang, GB Gao, MF Chang; IEEE BCTM; 9/1999 KK Lee, D An, JK Rhee, HC Park, YS Toon ; IEEE TENCON; 9/1999 I Takenaka, K Ishikura, H Takahashi, K Asano, J Morikawa, K Satou, K Kishi, K Hasegawa, K Tokunaga, F Emori, M Kuzuhara ; IEEE J SS Circ; 9/1999 M Yu, M Matloubian, P Petre, LR Hamilton, R Bowen, M Lui, HC Sun, CM Ngo, P Janke, DW Baker, RS Robertson; IEEE J SS Circ; 9/1999 TJ Jenkins, L Kehias, P Parikh, J Ibbetson, U Mishra, D Docter, M Le, J Pusl, D Widman ; IEEE J SS Circ; 9/1999 JM Nam, JW Ho, CM Rea, LY Hyun ; IEEE TENCON; 9/1999 D Ngo, WM Huang, JM Ford, D Spooner ; IEEE SOI Conf; 10/1999 K Eisenbeiser, D Droopad, JH Huang ; IEEE EDL; 10/1999 CY Hang, WR Deal, YX Qian, T Itoh ; IEEE Elec Letters; 10/1999 JCM Hwang, LT Kehias, JA Cook, MC Calcatera, ST Sheppard ; IEEE GaAs IC; 10/1999 LW Yang, KW Kobayashi, DC Streit, AK Oki, HC Yen, PC Grossman, TR Block, LT Tran, A GuiterrezAitken, LG Callejo, J Macek, S Maas ; IEEE GaAs IC; 10/1999 L Loval, A Darbandi, H Buret, M Soulard, F Michard ; IEEE GaAs IC; 10/1999 K Krishnamurthy, MJW Rodwell, SI Long ; IEEE GaAs IC; 10/1999 T Sasaki, Y Takada, Y Tanabe, T Nitta, Y Kakiuchi, M Yoshimura, K Fujieda, T Suzuki, H Kayano, M Hirose, Y Kitaura ; IEEE GaAs IC; 10/1999 CF Campbell, SA Brown ; IEEE GaAs IC; 10/1999 SMJ Liu, OSA Tang, W Kong, K NIchols, J Heaton, PC Chao ; IEEE GaAs IC; 10/1999 YC Chen, DL Ingram, D Yamauchi, B Brunner, J Kraus, M Barsky, R Grundbacher, SK Cha, R Lai, T Block, M Wojtowicz, TP Chin, B Allen, HC Yen, DC Streit ; IEEE GaAs IC; 10/1999 K Matsunaga, I Miura, N Iwata; IEEE GaAs IC; 10/1999 A Tessmann, L Verweyen, M Neumann, H Massler, WH Haydl, A Huelsmann, M Schlechtweg ; IEEE GaAs IC; 10/1999 H Kondoh, K Sekine, S Takatani, K Takano, H Kuroda, R Dabkowski ; IEEE GaAs IC; 10/1999 G Hau, TB Nishimura, N Iwata; IEEE GaAs IC; 10/1999 GK Wong, SI Long ; IEEE GaAs IC; 10/1999 H Ishida, T Yokoyama, H Furukawa, T Tanaka, M Maeda, S Morimoto, Y Ota, D ueda, C Hamaguchi; IEEE IEDM; 12/1999 KW Kobayashi, L Yang, A Guiterrez Aitken, E Kaneshiro, FM Yamada, HC Yen, TR Block, AK Oki, DC Streit; IEEE IEDM; 12/1999 AP Long, GD Edwards, JR Suffolk, WA Phillips, SD Wadsworth ; IEEE APMW Conf; 12/1999 W Simbuerger, HD Wohlmuth, P Weger, A Heinz ; IEEE J SS Circ; 12/1999 NS Cheng, PC Jia, DB Rensch, RA York ; IEEE Trans MTT; 12/1999 Y Hoshino, M Morikawa, S Kamohara, M Kawakami, T Fujioka, Y Matsunaga, Y Kusakari, S Ikeda, I Yoshida, S Shimizu; IEEE IEDM; 12/1999 JS Lim, SC Kang, SW Nam; IEEE MW Conference; 12/1999 J Cao, R Singh, B Liang, XW Wang, KE Sze, H Nakamura ; IEEE MW Conference; 12/1999 Y Tkachenko, Y Zhao, A Klimashov, CJ Wei, D Bartle ; IEEE MW Conference; 12/1999 TH Ng, BL Ooi, MS Leong, JY Ma, HS Ong, ST Chew ; IEEE MW Conference; 12/1999 AL Martin, A Mortazawi ; IEEE Trans MTT; 1/2000 M Kumar, Y Tan, J Sin, LX Shi, J Lau; IEEE ISSCC; 1/2000 J Carroll, R Flynt, S Brown ; IEEE ETS; 4/2000 JN Burghartz, M Hargrove, CS Webster, RA Groves, M Keene, KA Jenkins, R Logan, E Nowak ; IEEE TED; 4/2000 P Colantonio, F Giannini, E Limiti, G Saggio; IEEE Elec Letters; 4/2000 C Pobanz, M Matloubian, V Radisic, G Raghavan, M Case, M Micovic, M Hu, C Nguyen, S Weinreb, L Samoska; IEEE IPRM; 4/2000 131 [637] [638] [639] [640] [641] [642] [643] [644] [645] [646] [647] [648] [649] [650] [651] [652] [653] [654] [655] [656] [657] [658] [659] [660] [661] [662] [663] [664] [665] [666] [667] [668] JR Guthrie, M urteaga, D Scott, D Mensa, T Mathew, Q Lee, S Krishnan, S Jagananthan, Y Betser, MJW Rodwell ; IEEE IPRM; 4/2000 KW Kobayashi, AK Oki, LW Yang, A Guiterrez Aitken, P Chin, D Sawdai, W Okamura, J Lester, E Kaneshiro, PC Grossman, K Sato, TR Block, HC Yen, DC Streit ; IEEE IPRM; 4/2000 D Cui, D Sawdai, D pavlidis, SSH Hsu, P Chin, T Block ; IEEE IPRM; 4/2000 H Wang, L Samoska, T Gaier, A Peralta, HH Liao, YC Leong, S Weinreb, YC Chen, M Nishimoto, R Lai; IEEE MTT-S; 6/2000 J Cao, XW Wang, CK Quek, R Singh, H Nakamura ; IEEE MTT-S; 6/2000 HJ Siweris, A Werthof, H Tischer, T Grave, H Werthmann, RH Rasshofer, W Kellner; IEEE MTT-S; 6/2000 CP McCarroll, GD Alley, S Yates, R Matreci ; IEEE MTT-S; 6/2000 KI Jeon, JH Lee, SW Paek, DW Kim, WS Lee, CR Lim, HY Cha, HK Choi, KW Chung ; IEEE MTT-S; 6/2000 YJE Chen, M Hamai, D Heo, A Sutono, S Yoo, J Laskar ; IEEE MTT-S; 6/2000 T Satoh, T Shimura, S Ichikawa, A Betti-Berutto, C Poledrelli, Y Furukawa, Y Hasegawa, S Kuroda, J Fukaya ; IEEE MTT-S; 6/2000 JH Jeong, YW Kwon, SY lee, CY Cheon, EA Sovero ; IEEE MTT-S; 6/2000 J Staudinger, B Gilsdorf, D Newman, G Norris, G Sadowniczak, R Sherman, T Quach ; IEEE MTT-S; 6/2000 YC Leong, S Weinreb ; IEEE MTT-S; 6/2000 JJ Xu, S Keller, G Parish, S Heikman, YK Mishra, RA York ; IEEE MTT-S; 6/2000 YF Wu, D Kapolnek, JB Ibbetson, P Parikh, BP Keller, UK Mishra ; IEEE MTT-S; 6/2000 ST Sheppard, WL Pribble, DT Emerson, Z Ring, RP Smith, ST Allen, JW Palmour ; IEEE DRC; 6/2000 EY Chang, ES Fuh, CC Meng, KB Wang, SH Chen; IEEE TED; 6/2000 YW Kwon, KJ Kim, A Sovero, DS Deakin; IEEE Trans MTT; 6/2000 A Giry, JM Fournier, M Pons ; IEEE MTT-S; 6/2000 JP Fraysse, JP Viaud, M Campovecchio, P Auxemery, R Quere ; IEEE MTT-S; 6/2000 CY Hang, WR Deal, YX Qian, T Itoh ; IEEE MTT-S; 6/2000 S Tanaka, S Yamanouchi, Y Amamiya, K Hosoya, H Shimawaki, K Honjo; IEEE MTT-S; 6/2000 N Escalera, W Boger, P Denisuk, J Dobosz ; IEEE MTT-S; 6/2000 G Hau, TB Nishimura, N Iwata ; IEEE MTT-S; 6/2000 K Inoue, K Ebihara, H Haematsu, T Igarashi, H Takahashi, J Fukaya ; IEEE MTT-S; 6/2000 W Simbuerger, A Heinz, HD Wohlmuth, J Boeck, K Aufinger, M Rest ; IEEE MTT-S; 6/2000 XD Zhang, C Saycocie, S Munro, G Henderson; IEEE MTT-S; 6/2000 MJ Matilainen, KLI Nummila, EA Jaervinen, SJK Kalajo ; IEEE MTT-S; 6/2000 K Mori, S Shinjo, F Kitabayashi, A Ohta, Y Ikeda, 0 Ishida ; IEEE MTT-S; 6/2000 T Iwai, K Kobayashi, Y Nakasha, T Miyashita, S Ohara, K Joshin ; IEEE MTT-S; 6/2000 T Murae, K Fujii, T Matsuno ; IEEE MTT-S; 6/2000 SLG Chu, A Platzker, M Borkowski, R Mallavarpu, M Snow, A Bowlby, D Teeter, T Kazior, K Alavi; IEEE MTT-S; 6/2000 [669] [670] [671] [672] [673] [674] [675] [676] [677] [678] [679] [680] DL Ingram, YC Chen, I Stones, D Yamauchi, B Brunner, P Huang, M Biedenbender, J Elliott, R Lai, DC Streit, KF Lau, HC Yen ; IEEE MTT-S; 6/2000 AL Martin, A Mortazawi ; IEEE MTT-S; 6/2000 CS Yoo, QT Huang ; IEEE VLSI; 7/2000 RA Sadler, ST Allen, WL Probble, TS Alcorn, JJ Sumakeris, JW Palmour ; IEEE Cornell Conf on HPD; 7/2000 N Matsuno, H Yano, Y Suzuki, T Watanabe, S Tsubaki, T Toda, K Honjo ; IEEE Trans MTT; 7/2000 K Yamamoto, S Suzuki, K Mori, T Asada, T Okuda, A Inoue, T Miura, R Hattori, M Yamanouchi, T Shimura ; IEEE J SS Circ; 7/2000 BM Green, KK Chu, JA Smart, V Tilak, HT Kim, JR Shealy, LF Eastman ; IEEE MWGW Letters; 7/2000 E Chen, D Heo, M Hamai, J Laskar, D Bien ; IEEE RAWCON; 9/2000 HGA Huizing, F van Rijs, PHC Magnee, DMH Hartskeerl ; IEEE BCTM; 9/2000 EC Glass, JH Huang, J Staudinger, M Shields, MJ Martinez, OL Hartin, W Valentine, E Lan ; IEEE J SS Circ; 9/2000 K Krishnamurthy, R Vetury, S Keller, U Mishra, MJW Rodwell, SI Long; IEEE J SS Circ; 9/2000 K Matsunaga, I Miura, N Iwata ; IEEE J SS Circ; 9/2000 132 [681] [682] [683] [684] [685] [686] [687] [688] [689] [690] [691] [692] [693] [694] [695] [696] [697] [698] [699] [700] [701] [702] [703] [704] [705] [706] [707] [708] [709] [710] [711] [712] [713] [714] [715] [716] [717] [718] [719] [720] [721] [722] [723] [724] OSA Tang, SMJ Liu, PC Chao, WMT Kong, KC Hwang, K Nichols, J Heaton; IEEE J SS Circ; 9/2000 CS Whelan, PF Marsh, WE Hoke, RA McTaggart, PS Lyman, PJ Lemonias, SM Lardizabal, RE Leoni, SJ Lichwala, TE Kazior ; IEEE J SS Circ; 9/2000 PD Tseng, LY Zhang, GB Gao, MF Chang; IEEE J SS Circ; 9/2000 XN Zhu, XJ Chen, JT Ling ; IEEE ICMMT; 9/2000 TS Chen, Y Shen, FX Li, XJ Chen, Nanjing Uni; IEEE ICMMT; 9/2000 Y Tan, M Kumar, JKO Sin, LX Shi, J Lau; IEEE J SS Circ; 10/2000 E Alekseev, D Pavlidis, NX Nguyen, C Nguyen, DE Grider ; IEEE J SS Circ; 10/2000 A Sutono, D Heo, E Chen, K Lim, J Laskar ; IEEE Conf El Perf of Electronic Packaging; 10/2000 H Kawamura, K Sakuno, T Hasegawa, M Hasegawa, H Koh, H Sato; IEEE GaAs IC; 10/2000 T Moriuchi, W Abey, R Hajji, W Kennan, T Nakamura, S Maruyama, T Kitawada, Y Nonaka, E Mitani; IEEE GaAs IC; 10/2000 WCB Peatman, 0 Hartin, B Knappenberger, M Miller, R Hooper; IEEE GaAs IC; 10/2000 Y Butel, T Adam, B Cogo, M Soulard ; IEEE GaAs IC; 10/2000 W Okamura, LW Yang, A Guiterrez Aitken, E Kaneshiro, J Lester, D Sawdai, PC Grossman, K Kobayashi, HC Yen, A Oki, P Chin, T Block ; IEEE GaAs IC; 10/2000 SA Brown, JM Carroll ; IEEE GaAs IC; 10/2000 WMT Kong, SC Wang, PC Chao, DW Tu, K Hwang, OSA Tang, SM Liu, P Ho, K Nichols, J Heaton; IEEE EDL; 10/2000 K Krishnamurthy, S Keller, UK Mishra, MJW Rodwell, SI Long ; IEEE GaAs IC; 10/2000 A Bessemoulin, H Massler, A Huelsmann, M Schlechtweg ; IEEE MWGW Letters; 12/2000 YF Wu, PM Chavarkar, M Moore, P Parikh, BP Keller, UK Mishra ; IEEE IEDM; 12/2000 A Darbandi, H Buret, F Lavielle, M Peschoud ; IEEE Sem on RF PA; 12/2000 P Asbeck, C Fallesen ; IEEE ICECS; 12/2000 BX Gao, Y Wang; IEEE APMW Conf; 12/2000 T Tokumitsu; IEEE APMW Conf; 12/2000 T Ohguru, M Saito, E Morifuji, Koji Murakami, K Matsuzaki, T Yoshitomi, T Morimoto, HS Momose, Y Katsumata, H Iwai ; IEEE TED; 12/2000 K Mori, S Shinjo, F Kitabayashi, A Ohta, Y Ikeda, 0 Ishida ; IEEE Trans MTT; 12/2000 K Matsunaga, K Ishikura, I Takenaka, W Contrata, A Wakejima, K Ota, M Kanamori, M Kuzuhara ; IEEE IEDM; 12/2000 EY Chang, DH Lee, SH Chen ; IEEE APMC; 12/2000 MS Uhm, DP Jang, IB Yom, SP Lee; IEEE APMC; 12/2000 Y Trachenko, Y Zhao, C Wei, A Klimashov, D Bartle ; IEEE APMC; 12/2000 WY Kim, MC Chung, KH Lee, YG Yang, SH Kang, BM Kim ; IEEE APMC; 12/2000 BS Virdee, AS Virdee ; IEEE APMC; 12/2000 JY Lee, SG Han, JC Lee, B Lee, JH Kim, NY Kim, KI Chun ; IEEE APMC; 12/2000 HK Choi, JC Lee, B lee, JH Kim, NY Kim, US Hong ; IEEE APMC; 12/2000 G Hau, TB Nishimura, N Iwata ; IEEE MWWC Letters; 1/2001 C Fallesen, P Asbeck ; IEEE ISSCC; 1/2001 T Sowlati, D Leenaerts ; IEEE ISSCC; 1/2001 R Gupta, BM Ballweber, DJ Allstot ; IEEE J SS Circ; 1/2001 G Simin, X Hu, N Ilinskaya, J Zhang, A Tarakji, A Kumar, J Yang, M Asif Khan, R Gaska, MS Shur; IEEE EDL; 1/2001 TC Kuo, BB Lusignan; IEEE ISSCC; 1/2001 A Shirvani, DK Su, BA Wooley ; IEEE ISSCC; 1/2001 EM Chumbes, AT Schremer, JA Smart, Y Wang, NC MacDonald, D Hogue, JJ Komiak, SJ Lichwalla, RE Leoni, JR Shealy ; IEEE TED; 3/2001 SC Binari, K Ikossi, JA Roussos, W Kruppa, D Park, HB Dietrich, DD Koleske, AE Wickenden, RL Henry ; IEEE TED; 3/2001 LF Eastman, V Tilak, J Smart, BM Green, EM Chumbes, R Dimitrov, HT Kim, OS Ambacher, N Weimann, T Prunty, M Murphy, WJ Schaff, JR Shealy ; IEEE TED; 3/2001 SY Lee, BA Cetiner, H Torpi, SJ Cai, J Li, K Alt, YL Chen, CP Wen, KL Wang, T Itoh ; IEEE TED; 3/2001 M Micovic, A Kurdoghlian, P Janke, P Hashimoto, DWS Wong, JS Moon, L McCray, C Nguyen ; IEEE TED; 3/2001 133 [725] [726] [727] [728] [729] [730] [731] [732] [733] [734] [735] [736] [737] [738] [739] [740] [741] [742] [743] [744] [745] [746] [747] [748] [749] [750] [751] [752] [753] [754] [755] [756] [757] [758] [759] [760] [761] [762] [763] [764] [765] [766] [767] [768] [769] [770] [771] M Touirat, M Roger, JC Pesant, S Ajram, Y Crosnier, G Salmer ; IEEE EDL; 4/2001 SC Yang, HC Chiu, FT Chien, YJ Chan, JM Kuo ; IEEE EDL; 4/2001 YM Hsin, CH Lin, CC Fan, ST Su, MHT Yang, JCH Huang, KC Lin ; IEEE IPRM; 4/2001 I Aoki, SD Kee, D Rutlede, A Hajimiri ; IEEE CICC; 4/2001 CS Yoo, QT Huang ; IEEE J SS Circ; 4/2001 T Quach, W Okamura, A Guiterrez Aitken, T Jenkins, E Kaneshiro, L Kehias, A Oki, D Sawdai, P Watson, R Welch, R Worley, HC Yen ; IEEE IPRM; 4/2001 TB Nishimura, M Tanomura, K Azuma, K Nakai, Y Hasegawa, H Shimawaki ; IEEE MTT-S; 6/2001 W Bischof, M Alles, S gerlach, A Kruck, A Schueppen, J Sinderhauf, HJ Wassener; IEEE MTT-S; 6/2001 W Schuchter, G Krasser, V Schultheiss, G Hofer; IEEE MTT-S; 6/2001 D Gerna, A Giry, D Manstretta, D Belot, D Pache ; IEEE MTT-S; 6/2001 A Raghavan, E Gebara, CH Lee, S Chakraborty, D Mukherjee, J Battacharjee, D Heo, J Laskar ; IEEE MTT-S; 6/2001 F Temcamani, P Pouvil, 0 Noblanc, C Brylinski, P Bannelier, B Darges, JP Prigent ; IEEE MTT-S; 6/2001 E Ebihara, K Inoue, H Haematsu, F Yamaki, H Takahashi, J Fukaya ; IEEE MTT-S; 6/2001 T Murae, K Fujii, T Matsuno ; IEEE MTT-S; 6/2001 CH Lee, A Sutono, S Han, J Laskar ; IEEE MTT-S; 6/2001 Y Sasaki, H Kurusu, H Hoshi, T Hisaka, Y Mitsui ; IEEE MTT-S; 6/2001 J Estes, P Piel, G Shapiro, A Pavio, M Hurst, J Call, G Funk ; IEEE MTT-S; 6/2001 MP van der Heijden, HC de Graaf, LCN de Vreede, JR Gajadharsing, JN Burghartz ; IEEE MTT-S; 6/2001 H Darabi, S Khorram, HM Chien, MA Pan, S Wu, S Moloudi, JC Leete, JJ Rael, M Syed, R Lee, B Ibrahim, M Rofougaran, A Rofougaran ; IEEE J SS Circ; 6/2001 S Matsumoto, Y Hiraoka, T Sakai; IEEE TED; 6/2001 D Barataud, M Campovecchio, JM Nebus ; IEEE Trans MTT; 6/2001 CY Hang, WR Deal, YX Qian, T Itoh ; IEEE Trans MTT; 6/2001 H Okazaki, T Ohira, K Araki ; IEEE Trans MTT; 6/2001 G Hau, TB Nishimura, N Iwata ; IEEE Trans MTT; 6/2001 J Caldinhas Vaz, J Costa Freire ; IEEE Trans MTT; 6/2001 D Heo, A Sutono, E Chen, Y Suh, J Laskar ; IEEE MWWC Letters; 6/2001 P Colantonio, F Giannini, G Leuzzi, E Limiti ; IEEE MTT-S; 6/2001 YE Chen, YK Yoon, J Laskar, M Allen; IEEE MTT-S; 6/2001 R Emrick; IEEE MTT-S; 6/2001 R Purpas, F Brunner, R Doerner, B Janke, P Heymann, A Maasdorf, W Doser, P Auxemery, H Blanck, D Pons, J Wuerfl, W Heinrich ; IEEE MTT-S; 6/2001 I Takenaka, K Ishikura, H Takahashi, K Kishi, Y ogasawara, K Hasegawa, H takahashi, F Emori, N Iwata; IEEE MTT-S; 6/2001 C Fallesen, P Asbeck ; IEEE MTT-S; 6/2001 D Heo, A Sutono, E Chen, E Gebara, S Yoo, Y Suh, J Laskar, E Dalton, EM Tentzeris ; IEEE MTT-S; 6/2001 W Abey, T Moriuchi, R Hajii, T Nakamura, Y Nonaka, E Mitani, W Kennan, H Dang ; IEEE MTT-S; 6/2001 SY Zhang, J Cao, R Mcmorrow ; IEEE MTT-S; 6/2001 M Iwamoto, A Williams, PF Chen, A Metzger, CZ Wang, LE Larson, PM Asbeck; IEEE MTT-S; 6/2001 S Bousnina, FM Ghannouchi ; IEEE MTT-S; 6/2001 H Kobayashi, J Hinrichs, PM Asbeck ; IEEE MTT-S; 6/2001 C Duperrier, M Campovecchio, L Russel, M Lajugie, R Quere ; IEEE MTT-S; 6/2001 M Salib, HK Hahn, J Kositz, J Zingaro, A Ezis, A Gupta ; IEEE MTT-S; 6/2001 CY Hang, YX Qian, T Itoh ; IEEE MTT-S; 6/2001 E McShane, K Shenai, SK Leong ; IEEE MTT-S; 6/2001 BM Green, V Tilak, SJ Lee, HT Kim, JA Smart, KJ Webb, JR Shealy, LF Eastman ; IEEE MTT-S; 6/2001 F Fortes, MJ do Rosario ; IEEE Trans MTT; 7/2001 S Matsumoto, Y Hiraoka, T Sakai, T Tachi, T Ishiyama, T Kosugi, H Kamitsuna, M Murahuchi ; IEEE TED; 7/2001 M Kim, JB Hacker, RE Mihailovich, JF DeNatale ; IEEE MWWC Letters; 7/2001 ZQ Ma, S Mohammadi, LH Lu, P Bhattacharya, LPB Katehi, SA Alterovitz, GE Ponchak ; IEEE MWWC Letters; 7/2001 134 [772] [773] [774] [775] [776] [777] [778] [779] [780] [781] [782] [783] [784] [785] [786] [787] [788] [789] [790] [791] [792] [793] [794] [795] [796] [797] [798] [799] [800] [801] [802] [803] [804] [805] [806] [807] [808] [809] [810] [811] [812] [813] [814] [815] [816] [817] S Chakraborty, CH Lee, S Yoo, D Heo, A Raghavan, D Mukherjee, J Bhattacharjee, J Laskar ; IEEE RAWCON; 7/2001 JH Kim, B Palmer, R Harjani ; IEEE RAWCON; 7/2001 T Bito, T Kato, N Iwata ; IEEE TED; 7/2001 K Yamamoto, T Heima, A Furukawa, M Ono, Y Hashizume, H Komurasaki, S Maeda, H Sato, N Kato; IEEE J SS Circ; 7/2001 CC Hsiao, CW Kuo, YJ Chan; IEEE RAWCON; 7/2001 YS Noh, CS Park; IEEE RAWCON; 7/2001 MH Hella, M Ismail ; IEEE MWSCAS; 7/2001 P Baureis, M Peter, H Hein, F Oehler ; IEEE SMIC RF; 9/2001 JR Shealy ; IEEE SMIC RF; 9/2001 ZQ Ma, S Mohammadi, P Bhattacharya, LPB Katehi, SA Alterovitz, GE Ponchak ; IEEE SMIC RF; 9/2001 S Luo, T Sowlati ; IEEE Trans MTT; 9/2001 L Kehias, T Jenkins, T Quach, P Watson, R Welch, R Worley, AK Oki, HC Yen, A Guiterrez Aitken, W Okamura, E ]Kaneshiro ; IEEE MWWC Letters; 9/2001 JW Lee, KJ Webb ; IEEE MWWC Letters; 9/2001 Y Chung, S Cai, W Lee, Y Lin, CP Wen, KL Wang, T Itoh ; IEEE Elect Letters; 9/2001 T Senju, T Asano, H Ishimura ; IEEE BCIES; 9/2001 S Lam, WH ]Ki, MS Chan ; IEEE SOI Conf; 10/2001 YC Lee, CS ]Park; IEEE SSICT; 10/2001 Z Feng, XB Gao, CQ Wang; IEEE SSICT; 10/2001 HC Chiu, SC Yang, YJ Chan ; IEEE TED; 10/2001 A Sutono, DH Heo, YJE Chen, J Laskar ; IEEE Trans MTT; 10/2001 P Savary, A Girardot, G montoriol, F Dupis, B Thibaud, R Jaoui, L Chapoux, V Esnault, L Cornibert, O Izumi, D Hill, M Sadaka, H Henry, E Yu, M Tutt, M Majerus, R Uscola, F Clayton, C Rampley, S Klingbeil, K Rajagopalan, A Mitra, A Reyes ; IEEE GaAs IC; 10/2001 R Lai, R Grundbacher, M Barsky, A Oki, M Siddiqui, B Pitman, R Katz, P Tran, L Callejo, D Streit ; IEEE GaAs IC; 10/2001 T Quach, P Watson, W Okamura, E Kaneshiro, A Guiterrez Aitken, T Block, J Eldredge, T Jenkins, L Kehias, A Oki, D Sawdai, R Welch, R Worley ; IEEE GaAs IC; 10/2001 T Tokumitsu ; IEEE Trans MTT; 10/2001 T Takagi, K Yamauchi, Y Itoh, S Urasaki, M Komaru, Y Mitsui, H Nakaguro, Y Kazekami ; IEEE Trans MTT; 10/2001 T Hirayama, N Matsuno, M Fujii, H Hida ; IEEE GaAs IC; 10/2001 P Blount, J Cuggino, J McPhee ; IEEE GaAs IC; 10/2001 R Tayrani ; IEEE GaAs IC; 10/2001 JH Yoon, CH Seo ; IEEE MWWC Letters; 10/2001 CS Kim, JW Park, HK Yu ; IEEE IEDM; 12/2001 JW Lee, BM Green, V Tilak, SJ Lee, JR Shealy, LF Eastman, KJ Webb ; IEEE SDRC; 12/2001 W Walthes, A Pascht, M Berroth ; IEEE SDRC; 12/2001 DC Streit, D Sawdai, R Grunbacher, R Tsai, R Lai, A Guiterrez Aitken, A Oki ; IEEE APMC; 12/2001 S Chaki, T Hisaka, T Hirai, Y Sasaki, S Sakamoto, Y Mitsui ; IEEE APMC; 12/2001 S Halder, ZY Xiong, GI Ng, H Wang, HQ Zheng, K Radhakarishnan, JCM Hwang ; IEEE Trans MTT; 12/2001 VTS Vintola, MJ Matilainen, SJK Kalajo, EA Jaervinen ; IEEE Trans MTT; 12/2001 M Iwamoto, A Williams, PF Chen, AG Metzger, LL Larson, PM Asbeck ; IEEE Trans MTT; 12/2001 H Kobayashi, JM Hinrichs, PM Asbeck ; IEEE Trans MTT; 12/2001 BM Green, V Tilak, SJ Lee, HT Kim, JA Smart, KJ Webb, JR Shealy, LF Eastman ; IEEE Trans MTT; 12/2001 K Yamauchi, Y Iyama, M Yamaguchi, Y Ikeda, S Urasaki, T Takagi ; IEEE Trans MTT; 12/2001 SC Yang, HC Chiu, YJ Chan, HH Lin, JM Kuo ; IEEE TED; 12/2001 S Matsumoto, T Sakai ; IEEE TED; 12/2001 F Carrara, A Scuderi, G Tontodonato, G Palmisano ; IEEE IEDM; 12/2001 CW Kuo, CC Hsiao, YJ Chan; IEEE APMC; 12/2001 SH Chen, EY Chang, YC Lin ; IEEE APMC; 12/2001 HC Chiu, SC Yang, FT Chien, YJ Chan; IEEE EDL; 1/2002 135 [818] [819] [820] [821] [822] [823] [824] [825] [826] [827] [828] [829] [830] [831] [832] [833] [834] [835] [836] [837] [8381 [839] [840] [841] [842] [843] [844] [845] [846] [847] [848] [849] [850] [851] [852] [853] [854] [855] [856] [857] [858] [859] [860] KLR Mertens, MSJ Steyaert ; IEEE J SS Circ; 1/2002 YG Yang, YY Woo, BM Kim ; IEEE Trans MTT; 1/2002 JG Fiorenza, JA del Alamo ; IEEE TED; 4/2002 ZQ Ma, S Mohammadi, P Bhattacharya, LPB Katehi, SA Alterovitz, GE Ponchak; IEEE Trans MTT; 4/2002 HC Chiu, MJ Hwu, SC Yang, YJ Chan ; IEEE EDL; 4/2002 W Bakalski, W Simbuerger, D Kehrer, HD Wohlmuth, M Rest, AL Scholtz ; IEEE ISCAS; 4/2002 R Tayrani ; IEEE MTT-S; 6/2002 HC Huang, A Ezzeddine, A Darwish, B Hsu, J Williams, S peak ; IEEE MTT-S; 6/2002 T Matsuno, K Nishii, S Sonetaka, Y Toyoda, N Iwamoto; IEEE MTT-S; 6/2002 T Merkle, A Tessmann, S Ramberger ; IEEE MTT-S; 6/2002 EJ Crescenzi ; IEEE MTT-S; 6/2002 M Iwamoto, CP Hutchinson, JB Scott, TS Low, M Vaidyanathan, PM Asbeck, DC D'Avanzo ; IEEE MTTS; 6/2002 JM Schellenberg ; IEEE MTT-S; 6/2002 Y Wei, SM Lee, K Sundararajan, M Dahlstrom, M Urteaga, M Rodwell ; IEEE MTT-S; 6/2002 A Inoue, S Nakatsuka, R Hattori, Y Matsuda ; IEEE MTT-S; 6/2002 N Escalera, R Emrick, S Franson, B Farber, G Garrison, J Holmes, S Rockwell, B Bosco ; IEEE MTT-S; 6/2002 CH Lee, S Chakraborty, A Sutono, S Yoo, D Heo, J Laskar ; IEEE MTT-S; 6/2002 M Kaerkkaeinen, M Varonen, J Riska, P Kangaslahti, V Porra ; IEEE MTT-S; 6/2002 K Yamamoto, T Asada, S Suzuki, T Miura, A Inoue, S Miyakuni, J Otsuji, R Hattori, Y Miyazaki, T Shimura ; IEEE MTT-S; 6/2002 A Jitwin, 0 Bengtsson, J Olsson ; IEEE MTT-S; 6/2002 S Mohammadi, ZQ Ma, JH Park, P Bhattacharya, LPB Katehi, GE Ponchak, SA Alterovitz, KM Strohm, SF Luy ; IEEE MTT-S; 6/2002 MT Doan, Q Yin, KE Sze, PB Khannur, SC Rustagi, SJ Pang, D Foo, AB Ajjikuttira ; IEEE MTT-S; 6/2002 Y Mimino, K Nakamura, Y Hasegawa, Y Aoki, S Kuroda, T Tokumitsu ; IEEE MTT-S; 6/2002 K Fujii, M Adamski, P Bianco, D Gunyan, J Hall, R Kishimura, C Lesko, M Schefer, S Hessel, H Morkner, A Niedzwiecki ; IEEE MTT-S; 6/2002 WL Pribble, JW Palmour, ST Sheppard, RP Smith, ST Allen, TJ Smith, Z Ring, JJ Sumakeris, AW Saxler, JW Milligan ; IEEE MTT-S; 6/2002 A Shirvani, DK Su, BA Wooley ; IEEE J SS Circ; 6/2002 YK Chung, CY Hang, SJ Cai, YX Qian, CP Wen, KL Wang, T Itoh ; IEEE MTT-S; 6/2002 JP Fraysse, 0 Vendier, M Soulard, P Auxemery ; IEEE MTT-S; 6/2002 JM Schellenberg ; IEEE MTT-S; 6/2002 E Glass, M Shields, A Reyes ; IEEE MTT-S; 6/2002 J Staudinger, R Sherman, T Quach, M Miller, L Frye; IEEE MTT-S; 6/2002 S Kusunoki, K Yamamoto, T Hatsugai, K Tagami, H Nagaoka, N Tominaga, K Osawa, K Tanabe, S Sakurai, T lida ; IEEE MTT-S; 6/2002 L Zhao, A Pavio, B Stengel, B Thompson ; IEEE MTT-S; 6/2002 JJ Komiak, W Kong, K Nichols ; IEEE MTT-S; 6/2002 F Carrara, A Castorina, A Scuderi, G Palmisano ; IEEE MTT-S; 6/2002 A Raghavan, DH Heo, MK Maeng, A Sutono, KT Lim, J Laskar ; IEEE MTT-S; 6/2002 H Jaeger, A Grebennikov, E Heaney, R Weigel ; IEEE MTT-S; 6/2002 E Lan, E Johnson, B Knappenberger, M Miller ; IEEE MTT-S; 6/2002 RC Clarke, JW Palmour ; IEEE Proc IEEE; 6/2002 UK Mishra, P Parikh, YF Wu ; IEEE Proc IEEE; 6/2002 ST Sheppard, RP Smith, WL Pribble, Z Ring, T Smith, ST ALlen, J Milligan, JW Palmour ; IEEE DRC; 6/2002 A Aleeksov, M Kubovic, N Kaeb, U Spitzberg, I Daumiller, T Bauer, M Schreck, B Stritzker, E Kohn; IEEE DRC; 6/2002 A Agarwal, C Capell, B Phan, J Milligan, JW Palmour, J Stambaugh, H Bartlow, K Brewer ; IEEE HPD; 7/2002 136 [861] [862] [863] [864] [865] [866] [867] [868] [869] [870] [871] [872] [873] [874] [875] [876] [877] [878] [879] [880] [881] [882] [883] [884] [885] [886] [887] [888] [889] [890] [891] [892] [893] [894] [895] [896] [897] [898] [899] [900] [901] [902] [903] Y Wei, K Sundararajan, M Urteaga, Z Griffith, D Scott, V paidi, N Parthasarathy, M Rodwell ; IEEE HPD; 7/2002 KW Ho, HC Luong ; IEEE MWSCAS; 7/2002 S Lam, WH Ki, C Shen, PK Ko, M Chan; IEEE ICMMWT; 7/2002 JH Kim, JH Kim, YS Noh, CS Park ; IEEE ICMMWT; 7/2002 TK Quach, PM Watson, W Okamura, EN Kaneshiro, A Guiterrez Aitken, TR Block, JW Eldredge, TJ Jenkins, LT Kehias, AK Oki, D Sawdai, RJ Welch, RD Worley ; IEEE J SS Circ; 9/2002 ZQ Ma, S Mohammadi, P Bhattacharya, LPB Katebi, SA Alterovitz, GE Ponchak ; IEEE BCTM; 10/2002 MM Hella, M Ismail ; IEEE Proc CDS; 10/2002 M Tomaska, M Krnac, R Vazny ; IEEE ASDAM; 10/2002 HZ Liu, CC Wang, YH Wang, WL Huang, CH Chang, W Wu, CL Wu, CS Chang ; IEEE GaAs IC; 10/2002 KS Kong, D Boone, M King, B Nguyen, M Vernon, E Reese, G Brehm; IEEE GaAs IC; 10/2002 A Bessemoulin, J Dishong, G Clark, D White, P Quentin, H Thomas, D Geiger ; IEEE GaAs IC; 10/2002 J Dunn, G Freeman, D Harame, A Joseph, D Coolbaugh, R Groves, K Stein, R Volant, S Subbanna, VS Marangos, S St Onge, E Eshun, P Cooper, J Johnson, J Rieh, V Ramachandran, D Ahlgren, D Wang, X Wang; IEEE GaAs IC; 10/2002 JH Kim, JH Kim, YS Noh, YS Kim, SG Kim, CS Park ; IEEE GaAs IC; 10/2002 S Rockwell, R Emrick, B Bosco, S Franson, M Miller, E Johnson, J Crowder ; IEEE GaAs IC; 10/2002 A Bessemoulin, H Massler, R Quay, S Ramberger, M Schlechtweg ; IEEE GaAs IC; 10/2002 M Zargari, DK Su, CP Yue, S Rabii, D Weber, BJ Kaczynski, SS Mehta, K Singh, S Mendis, BA Wooley; IEEE J SS Circ; 12/2002 A Tessmann, S Kudszus, T Feltgen, M Riessle, C Sklarczyk, WH Haydl ; IEEE Trans MTT; 12/2002 YS Noh, JH Kim, JH Kim, CS Park; IEEE Electr Letters; 12/2002 HM Hsu, JG Su, CW Chen, DD Tang, CH Chen, YC Sun ; IEEE Trans MTT; 12/2002 R Kurpas, A Maassdorf, W Doser, P Heymann, B Janke, F Schnieder, H Blanck, P Auxemery, D Pons, W Heinrich, J Wuerfl ; IEEE IEDM; 12/2002 P Admane, M Patasani, B Viswanathan ; IEEE VLSI; 1/2003 S Sarkar, P Sen, A Raghavan, S Chakarborty, J Laskar ; IEEE VLSI; 1/2003 A Zolfaghari, B Razavi ; IEEE J SS Circ; 1/2003 CC Yen, HR Chuang ; IEEE MWWC Letters; 1/2003 BM Green, V Tilak, VS Kaper, JA Smart, JR Shealy, LF Eastman ; IEEE Trans MTT; 1/2003 V Paidi, SX Xie, R Coffie, B Moran, S Heikman, S Keller, A Chini, SP DenBaars, UK Mishra, S Long, MJW Rodwell ; IEEE Trans MTT; 1/2003 YK Chung, CY Hang, SJ Cai, YX Qian, CP Wen, KL Wang, T Itoh ; IEEE Trans MTT; 1/2003 W Nagy, J Brown, R Borges, S Singhal ; IEEE Trans MTT; 1/2003 IP Smorchkova, M Wojtowicz, R Sandhu, R Tsai, M Barsky, C Namba, PS Liu, R Dia, MD Truong, D Ko, J Wang, H Wang, A Khan ; IEEE Trans MTT; 1/2003 W Bakalski, N Ilkov, O Dernovsck, R Matz, W Simbuerger, P Weger, AL Scholz ; IEEE Elect Letters; 1/2003 N Vellas, C Gaquiere, A Minko, V Hoel, JC DeJaeger, Y Cordier, F Semond ; IEEE MWWC Letters; 3/2003 HY Tu, TH Chou, YS Lin, HC Chiu, PY Chen, WC Wu, SS Lu ; IEEE EDL; 3/2003 HY Chang, H Wang, M Yu, YH Shu ; IEEE MWWC Letters; 4/2003 F Villard, JP Prigent, E Morvan, C Dua, C Brylinski, F Temcamani, P Pouvil ; IEEE Trans MTT; 4/2003 H Komurasaki, T Sano, T Heima, K Yamamoto, H Wakada, I Yasui, M Ono, T Miwa, H Sato, T Miki, N Kato ; IEEE J SS Circ; 4/2003 TT Hung, MN ElGamal ; IEEE ISCAS; 4/2003 Y Ando, Y Okamoto, H Miyamoto, T Nakayama, T Inoue, M Kuzuhara ; IEEE EDL; 4/2003 KJ Herrick, CS Whealan, PF Marsh, SM Lardizabal ; IEEE IPRM; 4/2003 L Zhao, A Pavio, W Thompson ; IEEE MTT-S; 6/2003 J Udomoto, T Matsuzuka, S Chaki, K Kanaya, T Katoh, Y Notani, T Hisaka, T Oku, T Ishikawa, M Komaru, Y Matsuda ; IEEE MTT-S; 6/2003 C Dragon, W Brakensiek, D Burdeaux, W Burger, G Funk, M Hurst, D Rice; IEEE MTT-S; 6/2003 K Matsunaga, H Shimawaki ; IEEE MTT-S; 6/2003 M Morgan, S Weinreb ; IEEE MTT-S; 6/2003 137 [904] [905] [906] [907] [908] [909] [910] [911] [912] [913] [914] [915] [916] [917] [918] [919] [920] [921] [922] [923] [924] [925] [926] [927] [928] [929] [930] [931] [932] [933] [934] [935] [936] [937] [938] [939] [940] [941] [942] [943] [944] [945] [946] [947] [948] [949] AP Zhang, LB Rowland, EB Kaminsky, JW Kretchmer, V Tilak, AF Allen, BJ Edward ; IEEE MTT-S; 6/2003 F van Raay, R Quay, R Kiefer, M Schlechtweg, G Weimann ; IEEE MTT-S; 6/2003 HM Park, SH Cheon, JW Park, SC Hong ; IEEE MTT-S; 6/2003 NL Wang, C Dunnrowicz, X Chen, W Ma, HF Chau, X Sun, Y Chen, B Lein, IL Lo, CH Huang, MHT Yang; IEEE MTT-S; 6/2003 SQ Chen, E Reese, KS Kong ; IEEE MTT-S; 6/2003 K Fujii, H Morkner ; IEEE MTT-S; 6/2003 PS Wu, TW Huang, H Wang; IEEE MTT-S; 6/2003 M Varonen, M Kaerkkaeinen, P Kangaslahti, V Porra ; IEEE MTT-S; 6/2003 Y Wai, M Urteaga, Z Griffith, D Scott, SX Xie, V Paidi, N Parthasartathy, M Rodwell ; IEEE MTT-S; 6/2003 P Juurakko, V Saari, J Ryynaenen, K Halonen; IEEE MTT-S; 6/2003 PB Khannur; IEEE MTT-S; 6/2003 S Shinjo, HO Ueda, T Sugano, M Nakanishi, M Inoue, N Suematsu ; IEEE MTT-S; 6/2003 R Point, Z Li, W Foley, B Ingersoll, J Borelli, D Segarra, D Donoghue, C Liss, M Mendes, J Feigin, A Georgiadis, M Valery, E Dawe, D Losanno, R Quintal, M Nikitin, R Jabor, M Morin, KK O, G Dawe; IEEE MTT-S; 6/2003 M Ugajin, A Yamagishi, J Kodate, M Harada, T Tsukahara ; IEEE VLSI; 6/2003 K Tanaka, K Mochizuki, C Takubo, H matsumoto, T Tanoue, I Ohbu; IEEE MTT-S; 6/2003 JH Cha, YO Yang, BJ Shin, BM Kim ; IEEE MTT-S; 6/2003 KJ Herrick, SM Lardizabal, PF Marsh, CS Whelan ; IEEE MTT-S; 6/2003 H Brech, W Burger, C Dragon, B Pryor ; IEEE MTT-S; 6/2003 A Scuderi, F Carrara, A Castorina, G Palmisano ; IEEE MTT-S; 6/2003 Y Bito, T Ishigaki, H Shimawaki, Y Nashimoto ; IEEE MTT-S; 6/2003 T Niwa, T Ishigaki, H Shimawaki, Y Nashimoto; IEEE MTT-S; 6/2003 FY Colomb, A Platzker ; IEEE MTT-S; 6/2003 T Shimura, T Satoh, Y Hasegawa, J Fukaya ; IEEE MTT-S; 6/2003 unknown ; IEEE MTT-S; 6/2003 JH Kim, JH Kim, YS Noh, CS Park ; IEEE MTT-S; 6/2003 K Fujita, K Shirakawa, N Takahashi, Y Liu, T Oka, M Yamashita, K Sakuno, H Kawamura, M Hasegawa, H Koh, K Kagoshima, H Kijima, H Sato ; IEEE MTT-S; 6/2003 HC Chiu, TJ Yeh, SC Yang, MJ Hwu, YJ Chan ; IEEE TED; 6/2003 SY Zhang, P Brechko, J Mokoro, R McMorrow; IEEE MTT-S; 6/2003 N Tanzi, J Dykstra, K Hutchinson ; IEEE MTT-S; 6/2003 T Biondi, F Carrara, A Scuderi, G Palmisano ; IEEE MTT-S; 6/2003 I Aoki, S Kee, D Rutledge, A Hajimiri ; IEEE MTT-S; 6/2003 S Pajic, ZB Popovic ; IEEE Trans MTT; 7/2003 SX Xie, V Paidi, R Coffie, S Keller, S Heikman, B Moran, A Chini, SP DenBaars, U Mishra, S Long, MJW Rodwell ; IEEE MWWC Letters; 7/2003 HC Chiu, SC Yang, CK Lin, MJ Hwu, YJ Chan ; IEEE TED; 7/2003 A Keerti, A Pham ; IEEE Elect Letters; 7/2003 T Oka, K Fujita, K Shirakawa, N Takahashi, Y Liu, M Yamashita, H Kawamura, M Hasegawa, H Koh, H Kagoshima, H Kijima, K Sakuno ; IEEE ISCS; 7/2003 T Sowlati, DMW Leenaerts ; IEEE J SS Circ; 7/2003 JA O'Sullivan, C Delabie, KG McCarthy, A Murphy, PJ Murphy ; IEEE HFPSC; 9/2003 CK Chu, HK Huang, CC Wang, YH Wang, CC Hsu, W Wu, CL Wu, CS Chang ; IEEE ESSCIRC; 9/2003 W Bakalski, A Vasylyev, W Simbuerger, R Thueringer, HD Wohlmuth, AL Scholz, P Weger ; IEEE BCTM; 9/2003 A Wakejima, K Ota, K Matsunaga, M Kuzuhara ; IEEE TED; 9/2003 V Adivarahan, M Gaevski, WH Sun, H Fatima, A Koudymov, S Saygi, G Simin, J Yang, M Afir Khan, A Tarakji, MS Shur, R Gaska ; IEEE TED; 9/2003 A Bessemoulin, R Quay, S Ramberger, H Massler, M Schlechtweg ; IEEE J SS Circ; 9/2003 M Spirito, LCN de Vreede, LK Nanver, S Weber, JN Burghartz ; IEEE J SS Circ; 9/2003 R Negra, R Vogt, W Baechtold ; IEEE IMOC; 9/2003 S Hamedi Hagh, CAT Salama ; IEEE CICC; 9/2003 138 [950] [951] [952] [953] [954] [955] [956] [957] [958] [959] [960] [961] [962] [963] [964] [965] [966] [967] [968] [969] [970] [971] [972] [973] [974] [975] [976] [977] [978] [979] [980] [981] [982] [983] [984] [985] [986] [987] [988] [989] [990] [991] [992] [993] [994] [995] A Adahl, H Zirath ; IEEE EMC; 10/2003 K Seeman, S Ramberger, A Tessmann, R Quay, J Schneider, M Riessle, H Walcher, M Kuri, R Kiefer, M Schlechtweg ; IEEE EMC; 10/2003 S Bantas, Y Stratakos, N Kanakaris, Y Katsoulis, P Papadopoulos, M Margaras, V Korou, H Peyravi, Y Koutsoyannopoulos ; IEEE VTC; 10/2003 BP Yan, ES Yang, YF Yang, XQ Wang, CC Hsu; IEEE TED; 10/2003 C Lee, P Saunier, JW Yang, M Asif Khan ; IEEE EDL; 10/2003 Y Okamoto, Y Ando, K Hataya, H Miyamoto, T Nakayama, T Inoue, M Kuzuhara ; IEEE Elect Letters; 10/2003 N Ceylan, JE Mueller, T Pittorino, R Weigel ; IEEE EMC; 10/2003 N Le Gallou, JF Villemazet, B Cogo, JL Cazaux, A Mallet, L Lapierre ; IEEE EMC; 10/2003 M Park, HK Ahn, DM Kang, HG Ji, JK Mun, HC Kim, KI Cho ; IEEE EMC; 10/2003 N Srirattana, A Raghavan, D Heo, PE Allen, J Laskar ; IEEE EMC; 10/2003 JB Johnson, AJ Joseph, D Sheridan, RM Malladi ; IEEE GaAs IC; 10/2003 M Rancanelli, PX Ma, P Kempf; IEEE GaAs IC; 10/2003 S Handa, E Suematsu, H Tanaka, Y motouchi, M Yagura, A Yamada, H Sato ; IEEE GaAs IC; 10/2003 YH Wang, HZ Liu, HK Huang, CC Wang, CH Chang, W Wu, CL Wu, CS Chang ; IEEE EDMO; 10/2003 GH Jessen, RC Fitch, JK Gillespie, GD Via, NA Moser, MJ Yannuzzi, A Crespo, JS Sewell, RW Dettmer, TJ Jenkins, RF Davis, J Yang, M Asif Khan, SC Binari ; IEEE EDL; 10/2003 H Komizo, Y Arai, T Saito; IEEE MTT-S; 6/1977 JW Lee, LF Eastman, KJ Webb ; IEEE Trans MTT; 10/2003 PM Smith, D Dugas, K Chu, K Nichols, KG Duh, J Fisher, L MtPleasant, D Xu, L Gunter, A Vera, R Lender, D Meharry ; IEEE GaAs IC; 10/2003 N Saka, M Nakamura, M Shimada, T Kimura, H Motoyama, I Hase ; IEEE GaAs IC; 10/2003 AR Behzad, ZM Shi, SB Anand, L Lin, KA Carter, MS Kappes, TS Lin, T Nguyen, D Yuan, S Wu, YC Wong, V Fong, A Rofougaran ; IEEE J SS Circ; 12/2003 K Joshin, T Kikkawa, H Hayashi, T Maniwa, S Yokokawa, M Yokoyama, N Adachi, M takikawa ; IEEE IEDM; 12/2003 M Ferndahl, HO Vickes, H Zirath, I Angelov, F Ingvarson, A Litwin ; IEEE MWWC Letters; 12/2003 X Jiang, AL Martin, A Mortazawi ; IEEE MWWC Letters; 12/2003 HC Chiu, SC Yang, YJ Chan, SH Chen, WS Liu, JI Chyi ; IEEE TED; 1/2004 JW Lee, KJ Webb ; IEEE Trans MTT; 1/2004 B Sahu, GA Rincon-Mora ; IEEE Trans MTT; 1/2004 P Colantonio, F Giannini, E Limiti, V Teppati ; IEEE Trans MTT; 1/2004 W Wang, YP Zhang ; IEEE MWWC Letters; 1/2004 X Jiang, SC Ortiz, A Mortazawi ; IEEE Trans MTT; 1/2004 T Shimizu, Y Nunogawa, T Furuya, S Yamada, I Yoshida, M Hotta; IEEE ISSCC; 1/2004 CM Wang, HT Hsu, HC Shu, YM Hsin ; IEEE EDL; 1/2004 B Toner, R Dharmalinggam, VF Fusco ; IEEE MWAntProp; 1/2004 L Samoska, A Peralta, M Hu, M Micovic, A Schmitz ; IEEE MWWC Letters; 1/2004 TY Yum, Q Xue, CH Chan ; IEEE Trans MTT; 3/2004 A Fukuda, H Okazaki, T Hirota, Y Yamao ; IEEE MWWC Letters; 3/2004 CW Park ; IEEE Sarnoff AWWC; 4/2004 M Ugajin, A Yamagishi, J Kodate, M Harada, T Tsukahara ; IEEE J SS Circ; 4/2004 S Forestier, P Bouysse, R Quere, A Mallet, JM Nebus, L Lapierre ; IEEE Trans MTT; 4/2004 CC Huang, HT Pai, KY Chen ; IEEE Trans MTT; 4/2004 R Behtash, H Tobler, FJ Berlec, V Ziegler, H Leier, RS Balmer, T Martin, M Neuburger, H Schumacher; IEEE Elect Letters; 4/2004 YH Chee, J Rabaey, AM Niknejad; IEEE ISCAS; 4/2004 P Chen, HR Chang, X Li, B Luo; IEEE ISPSD; 4/2004 A Chini, D Buttari, R Coffie, L Shen, S Heikman, A Chakraborty, S Keller, UK Mishra ; IEEE EDL; 4/2004 MD Hampson, SC Shen, RS Schwindt, RK Price, U Chowdhury, MM Wong, TG Zhu, DW Yoo, RD Dupuis, M Feng ; IEEE EDL; 4/2004 CM Snowden ; IEEE Proc CDS; 6/2004 UR Pfeiffer, SK Reynolds, BA Floyd ; IEEE MTT-S; 6/2004 139 JX Deng, P Gudem, LE Larson, PM Asbeck ; IEEE MTT-S; 6/2004 K Choi, DJ Allstot, V Krishnamurthy ; IEEE MTT-S; 6/2004 Y Zhang, P Heydari ; IEEE MTT-S; 6/2004 W Bakalski, K Kitlinski, G Doning, B Kapfelsperger, W Oesterreicher, W Auchter, R Weigel, AL Scholz ; IEEE MTT-S; 6/2004 [1000] JG Fiorenza, J Scholvin, JA del Alamo ; IEEE EDL; 10/2003 [1001] ID Robertson, KKM Cheng, AH Aghvami ; IEEE ICC; 6/2004 [1002] HG Henry, G Augustine, GC DeSalvo, RC Brooks, RR Barron, JD Oliver, AW Morse, BW Veasel, PM Esker, RC Clarke; IEEE TED; 6/2004 [1003] YS Noh, CS Park; IEEE J SS Circ; 6/2004 [1004] S Piotrowicz, E Chartier, JC Jacquet, D Floriot, JM Coupat, C Framery, P Eudeline, P Auxemery ; IEEE MTT-S; 6/2004 [1005] YJ Jung, H Jeong, E Song, J Lee, SW Lee, DH Seo, IH Song, SH Jung, JB Park, DK Jeong, SI Chae, WC Kim; IEEE J SS Circ; 7/2004 [1006] JB Johnson, AJ Joseph, DC Sheridan, RM Maladi, PO Brandt, J Persson, J Andersson, A Bjorneklett, U petersson, F Abasi, L Tilly ; IEEE J SS Circ; 10/2004 [996] [997] [998] [999] 140