ULTRA LOW VOLTAGE CURRENT MULTIPLIER/DIVIDER Yngvar Berg and Tor Sverre Lande Dept. of Informatics, University of Oslo, Gaustadalleen 23, P.O.Box 1080, Blindern, N-0316 Oslo, Norway. Fax: +47 22 85 24 01. Email:yngvarb@ifi.uio.no 10 10 10 -7 Standard transistor -8 max I ds -9 Weak inversion model -10 [A] Exact current ds I. Introduction 10 I Abstract— In this paper we are presenting novel current multiplier/divider circuits utilizing the power of FGUVMOS transistors. Even for ultra low supply voltages a proper operation is achieved with threshold shifting using post-fab. tuning. Measured results are provided of a multiplier working over a large magnitude in current at 0.7V power supply. 10 10 10 -11 -12 I bec C1 -13 V As the current-mode approach to analog circuit design [1] is gaining interest due to better performance, another compelling reason for current-mode circuits is the decreasing power supply of digital microelectronics. The upside of sharing both silicon and power with digital electronics challenges the analog designer to come up with solutions. As the power supply is lowered, the available headroom is shrinking, so current-mode design techniques provides some kind of relief by demanding much less headroom (compressed voltage-mode). Looking into the future, ultra low supply voltage (ULV) Vdd < 1V is coming. Improved precision in production will permit a reduced threshold voltage with improved matching, but still the threshold voltage will “steal” an increasing part of the available headroom. One possible solution to the matching problem is post-fabrication tuning like laser-trimming. Techniques requiring individual tuning of each poweredup circuit are slow and expensive. Another approach to threshold matching is to use MOS transistors with capacitive coupling to a floating gate as indicated in figure 1. A change in the stored charge of the floating gate will effectively shift the threshold voltage seen from the capacitively coupled input terminal. The success of this approach is clearly dependent on both the accuracy and the overhead penalty in terms of silicon area and production cost. We have developed a novel and very simple approach [2], [3] to program the floating gate of MOStransistors, called FGUVMOS transistors, available in any double-poly CMOS process. By exposing the circuit to short-wave UV-light, the threshold voltage of both PMOS and NMOS transistors can be tuned simply by reverse-biasing the power-rails. In the following, we will apply FGUVMOS transistors to design a current-mode multiplier/divider circuit for ULV operation, but in order to understand the design, we have to explain some more details about FGUVMOS circuits in general. There 10 10 10 10 -14 C 2 V -15 -16 in fg I ds Cm min I ds -17 0 0.1 0.2 0.3 0.4 0.5 V in [V] Fig. 1. Multiple input n-channel Floating-gate UVMOS (FGUVMOS) transistor. is always one PMOS stacked on top of one NMOS transistor. The height is always two with a common drain-node, but of course additional MOS transistors may be added in parallel (both Ps and Ns). However, each floating gate may have several inputs connected through floating capacitors, compensating for the limited stacking. All FGUVMOS circuits must be tuned or programmed using short-wave UVlight (UV-C) [2], [4]. Note that all transistors on a chip, or even a wafer, can be programmed simultaneously without using any additional programming circuitry, except the power-rails. Two important conditions are established during the tuning process: 1. The threshold voltage of all PMOS transistors is typically programmed to a small negative voltage relative to the Vdd supply rail, while the threshold voltages of all NMOS transistors are typically tuned to a small positive voltage relative to the Vss supply rail (substrate). By reverse biasing the supplyrails, the source and drain of all transistors are interchanged. The nodes between the PMOS and NMOS are strongly driven by a source-follower configuration to a equilibrium voltage, Vbe (≡ VDD /2), maintained also under normal biasing conditions. The current flowing through the stacked transistors is matched and will result in a balanced equilibrium current, Ibec , under normal biasing. 2. A useful consequence of the current matching is the equalized transconductances [2], due to shifting of the threshold voltage. We have that is gmN /gmP = 1 in weak inversion and gmN /gmP = p µN /µP in strong inversion. In other words the difference in mobility between N-type and P-type transistors is compensated without altering transistor dimensions. The symmetry is another important property of FGUVMOS circuits and will compensate for systematic errors like the Early effect and mobility differences between PMOS and NMOS. Po I in1 I in2 C po I Po V po V in2 V in1 V no I No C i1 C i3 C no C i2 Ni1 No Ni2 II. FGUVMOS circuits In spite of the design restrictions imposed by the tuning scheme, it is possible to make powerful circuits. A crucial design parameter is the sizing of the coupling capacitors. For a multiple input FGUVMOS transistor each input has by design an effective coupling capacitance, Ci , to the floating gate. The input signal is attenuated with a factor ki = Ci /CT , where CT is the total load capacitance seen from the input. This load capacitance will increase with the number of inputs and may even be non-linear. For FGUVMOS circuits it is convenient to express the behavior as modulation of the equilibrium condition. For simplicity, we will model the weak inversion behavior knowing that a similar analysis may be done for strong inversion as well. The input modulation of the drain current as a function of the ith input terminal may be expressed as 1 exp{ nU (Vi − Vdd /2)ki }. The accumulated drain t current modulation of m inputs is expressed as the Qm 1 (Vi −Vdd /2)ki }. The effective product i=1 exp{ nU t drain current of a FGUVMOS transistor may then be written as Ids = Ibec m Y i=1 exp{ 1 (Vi − Vdd /2)ki }. nUt Furthermore, we may express the min and max currents in terms of the balanced equilibrium current and current modulation max Ids min Ids 1 = Ibec exp{ Vdd ki } 2nUt Pm = Ibec R i=1 ki Pm = Ibec R i=1 −ki Ibec = max , Ids 1 Vdd } and the maximum dywhere R = exp{ 2nU t P m namic range, assuming i=1 ki = 1, of the transis2 tor current is R . Although this is an expression for a m-input FGUVMOS NMOS transistor, the number of inputs are usually kept small in order to maintain a sufficient SNR. The expression for PMOS transistors are very similar. Fig. 2. Minch’ current divider (original without the PMOS). III. Current multipliers and dividers We are now equipped with the required tools for FGUVMOS circuit design. Although these kind of circuits are kind of exotic, the basic idea of the circuit in figure 2 is reported earlier in [5]. The input currents may be expressed as Iin1 Iin2 1 (Vin1 − Vdd /2)ki1 } nUt 1 = Ibec exp{ (Vin1 − Vdd /2)ki1 } nUt 1 · exp{ (Vin2 − Vdd /2)ki2 }. nUt = Ibec exp{ In our equilibrium analysis the output current IN o may be expressed as IN o 1 (Vin2 − Vdd /2)kno } nUt kno Iin2 ki2 = Ibec ( . (1) ki3 ki3 ) 1− (Ibec ) ki1 (Iin1 ) ki1 = Ibec exp{ If ki3 /ki1 = 1 equation 1 reduces to IN o = Ibec (Iin2 /Iin1 )kno /ki2 which demonstrates the dividing and scaling property of the circuit. For the output current of the PMOS transistor, IP o , we obtain IP o = Ibec ( Iin1 kkpo ) i2 . Iin2 (2) Again if kno /ki2 = kpo /ki2 equation 2 reduces to IP o = IN o giving us the inverse output. In fact the inverse of any FGUVMOS circuit may be obtained by adding the opposite type of MOS device. We have added the PMOS giving the inverted output. The simulated output currents of the current divider are shown in figure 3. The circuit behaves reasonably well within two order of magnitude. A feature of FGUVMOS circuits is the low supply voltages of only 0.7V in these simulations. The symmetry properties of FGUVMOS circuits may be used to change the current divider into a current multiplier. By flipping the Ni2 transistor −7 10 Ino, Iin2=Imax Po I in2 I in1 Ino, Iin2=Ibec C o3 −8 10 Iout Ipo, Iin2=Imax C o4 V in2 V in1 I Po V po V no Ino, Iin2=Imin I No C i1 −9 C i2 C o1 C o2 10 No Ipo, Iin2=Imax −10 10 Ni2 Ni1 Ipo, Iin2=Ibec −10 −9 10 −8 10 Fig. 6. Improved FGUVMOS current multiplier. −7 10 10 Iin1 −7 10 Fig. 3. Current divider characteristics. Pi I in1 Ipo, Iin2=Imin Ipo, Iin2=Ibec Po −8 10 I Po C po C i2 V po Ino, Iin2=Imax Iout C i3 V in2 V in1 Ipo, Iin2=Imax V no I No −9 10 C no C i1 Ino, Iin2=Ibec I in2 Ni No Ino, Iin2=Imin −10 10 Fig. 4. Current multiplier. −10 −9 10 −8 10 10 −7 10 Iin1 in figure 2 and make it a PMOS transistor, we get a current multiplier as shown in figure 4. The output current IP o may be written as IP o = Ibec ( (Iin1 ) ki3 ki1 Iin2 k 1+ ki3 (Ibec ) ) kpo ki2 . (3) i1 If ki3 /ki1 = 1 equation 3 reduces to IP o = Ibec (Iin2 Iin1 /(Ibec )2 )kpo /ki2 . Again we may read out an inverted version of our transfer function through the PMOS transistor. The simulated current multiplier characteristics are shown in figure 5 and are very similar to the current −7 10 Ino, Iin2=Imin Fig. 7. Improved FGUVMOS current multiplier characteristics. divider regarding dynamic range. IV. Improved multiplier/divider circuits A current multiplier with extended dynamic range and better symmetry may be implemented as a FGUVMOS circuit. By using two coupling capacitors to the output stages, the effective coupling from each input terminal is reduced giving an increased dynamic range. In addition we get symmetrical input stages improving linearity of the multiplier. The circuit is shown in figure 6. Applying our equilibrium analysis we find the output current, IN o , may be expressed as IN o Ipo, Iin2=Imax = Ibec ( Iin1 kko1 Iin2 kko2 ) i1 ( ) i2 . Ibec Ibec (4) −8 10 Iout Ino, Iin2=Ibec Ipo, Iin2=Ibec −9 10 Ino, Iin2=Imax Ipo, Iin2=Imin −10 10 −10 10 −9 −8 10 10 −7 10 Iin1 Fig. 5. Simulated output current of the current multiplier. Again if ko1 /ki1 = ko2 /ki2 we have that IN o = Ibec {(Iin2 Iin1 )/(Ibec )2 }ko1 /ki1 . In figure 7 the simulated output of the current multiplier is shown. As expected the dynamic range is increased to three orders of magnitude and might be increased further by capacitive division limited by the acceptable SNR. An interesting feature is to play with the terms rx and ry . If we let rx = ry = 1/2 in equation 4 we get p Iin1 Iin2 IN o = −5 10 Pi Po I n1 −6 10 C o4 C i2 C o3 I Po −7 10 V po V in2 −8 V in1 10 V no C i1 C o2 I in2 Iout (A) I No C o1 Ni −9 10 No −10 10 −11 10 Fig. 8. Improved FGUVMOS current divider. −12 10 −7 10 −13 10 Ipo, Iin2=Imax −9 10 −8 10 Iin (A) −7 10 −6 10 Fig. 10. Measured performance of the current multiplier. The output current as a function of the input current was measured for different multiplication factors (voltages). Ipo, Iin2=Ibec −8 10 Ino, Iin2=Imin Iout −10 10 Ipo, Iin2=Imin −9 10 Ino, Iin2=Ibec Ino, Iin2=Imax −10 10 −10 −9 10 −8 10 10 −7 10 Iin1 Fig. 9. Hspice simulation of the improved FGUVMOS current divider output. IP o = I2 √ bec . Iin1 Iin2 As illustrated, different compressive or expansive functions might be added just by changing capacitive ratios. To make the picture complete, we present the improved FGUVMOS current divider. The output current IN o in figure 8 may be expressed as IN o = Ibec ( Iin1 kko1 Ibec kko2 ) i1 ( ) i2 . Ibec Iin2 The measured performance of a current multiplier is shown in figure 10 with a supply voltage of 0.7V . The output current is measured as a function of the input current while varying the multiplying input voltage. Note that a pure current-mode version of this circuit would not achieve a dynamic range of this magnitude, but with one voltage-mode input terminal, we could exceed the 0.7V power supply voltage. For higher current-levels transition to strong inversion may be observed, but for weak inversion the linearity is exceptional. VI. Conclusion Novel current-mode multipliers and dividers have been proposed exploring the power of the FGUVMOS transistor. Low supply voltage is combined with minimum sized transistors to form a working current multiplier/divider circuit at 0.7V . The feasibility of our approach is proven through measured results of a test circuit implemented in standard double-poly CMOS. References If ko1 /ki1 = ko2 /ki2 we have that IN o = Ibec (Iin1 /Iin2 )ko1 /ki1 . The simulated output currents of the improved current divider are shown in figure 9. We have our dividing function with improved linearity and dynamic range. [1] V. Measured results [4] A slightly simplified version of the circuit in figure 7 was implemented in AMS 0.8µ CMOS using minimum transistors. One of the input-current branches was removed and replaced with a direct capacitive coupling to a NMOS output transistor. Before testing the circuits a tuning was done using a standard 4W UV-eraser according to the procedure required for FGUVMOS transistors [2], [3], [6]. [2] [3] [5] [6] Analogue IC Design: The Current-mode Approach, Edited book by Toumazou C. et. al., Peter Penegrinus, 1990 Y. Berg and T.S. Lande. Programmable FloatingGate Mos Logic for Low-Power Operation., ISCAS, june. 1997. Y. Berg and T.S. Lande. Low-Voltage Floating-Gate Current Mirrors., ASIC97, october. 1997. R.G. Benson and D.A. Kerns. UV-Activated Conductances Allow For Multiple Scale Learning., IEEE Transactions on Neural Networks, vol. 4, no. 3, may 1993. B.A. Minch, C. Diorio, P. Hasler and C.A. Mead. Translinear Circuits Using Subthreshold FloatingGate MOS Transistors., Analog Integrated Circuits and Signal Processing, 9, pp.167-179, Kluwer Academic Publishers, 1996. Y. Berg and T.S. Lande. Tunable Current Mirrors for Ultra Low Voltage., ISCAS99, may-june. 1999.