i University of Twente Faculty of Electrical Engineering, Mathematics & Computer Science A Low Power NMOS LDO in the Philips CO50PMU Process S.N.Easwaran MSc. Thesis March 2006 Supervisors: Dr. ir. R.A.R van der Zee Prof.dr. ir. B.Nauta Report number: 067.3144 Chair of Integrated Circuit Design Faculty of Electrical Engineering, Mathematics & Computer Science University of Twente P. O. Box 217 7500 AE Enschede The Netherlands Abstract In this report a new architecture of an NMOS Low Drop Out (LDO) Regulator is proposed. It features a new frequency compensation technique to enable better regulation. The LDO Regulator handles large load currents with good stability (25° phase margin for low load currents from no load till 2mA and then towards 90°for larger load currents) and a good transient response (overshoot and undershoot less than 3% for a load step from 1mA to 500mA). The quiescent current of the LDO is 50µA, with a load capacitor of 470nF. The LDO is scalable for different maximum load currents without impact on stability. In Standby mode the LDO is stable with a quiescent current three times smaller than the normal operation. The application of the NMOS LDO is in Power Management Units for all the handheld and mobile application devices like the mobile phones, Apple IPODs etc. iii Acknowledgement This thesis describes the result of a Master of Science assignment at Philips Semiconductors, Nijmegen, The Netherlands. This assignment has been carried out from August 2005 to March 2006 at the PMU-ABB (Power Management Unit-Analog Base Band) for completing the Master degree in Electrical Engineering in the section of ICD at the University of Twente, The Netherlands. I would like to thank Ir.Alberto van Burgh, Development Manager as he hired me to the PMU-ABB group which enabled me to move from Philips Semiconductors (PS), Zurich to PS, Nijmegen to continue my Analog Design Career. I would like to thank Ir.Ferdinand Sluijs, Innovation Manager and Ir. Johan.M. van der Wiel, Development Group Leader for their support and for offering me the NMOS LDO topic as my Master Thesis subject. I would like to thank Prof.Dr.ir. Bram Nauta (Head ICD, University Of Twente (UT)) for his wonderful support and he and his team were the backbone for my success during my Master Studies at the UT. At the outset I would like to thank Dr.ir.Ronan van der Zee for all his support, guidance and patience. As my supervisor, he has constantly enforced me to be in focus towards achieving this goal. His observations and comments helped me to establish the overall direction of the research and move forward with investigation in depth. I met him only once a week but his overall support with several exchange of e-mails and phone calls were at the best in order to help me carry out the NMOS LDO research and write the Thesis in a better manner. His way of work coupled with intuitive analysis helped me to understand the concepts in a better way. I greatly appreciate Ing. Max Martin, my supervisor at PS, Nijmegen for his whole hearted support and suggestions during the whole thesis work. As my supervisor at PS, Nijmegen he helped me to know more about the specifications of the NMOS LDO, its environment in the real application which helped me stay with in the specifications and limitations. I have no words to express his wonderful support to me during this research period. I would like to thank Dr.ir.K.J.de Langen, BU Automotives, PS,Nijmegen for his help and fruitful discussions during this research. I would like to thank my friends from Talent Telecom Solutions (supporting PMU-ABB group at PS, Nijmegen) for fruitful discussions to gather some information on the competitor LDOs in the market. I would like to thank other colleagues at PS, Nijmegen for the pleasant working environment. I would like to thank my fellow students at UT for the pleasant learning environment. I would like to thank my friends Balaji, Shankar and Vishnu for providing me a lot of support and enthusiasm during my stay at the UT campus. I acknowledge many friends in Enschede who have made my 1.5 year study so cheerful. For me, UT campus looked as a Home away from Home. Last, but not the least, I would like to dedicate this thesis to my family in India, for their love, patience, and understanding. iv List Of Abbreviations PMU - Power Management Unit BOM - Bill Of Materials LDO - Low Drop Out Regulator UGB - Unity Gain Band Width LHP - Left Half Plane RHP - Right Half Plane ESR - Equivalent Series Resistance ESL - Equivalent Series Inductance SMD - Surface Mounted Device v Table Of Symbols gm*- Transconductance of in Siemens (S) R*- Output Impedance in Ohms (Ω) A(s) - Open Loop Gain in decibels (dB) β - Feedback Factor ϕ - Phase Margin in degrees (°) CL- Load Capacitor fp* - Frequency Of The Pole in Hertz (Hz) fz* - Frequency Of The Zero in Hertz (Hz) fτ - Unity Gain Bandwidth in Hertz (Hz) µn - Mobility of the electron in m2/Vsec Cox – Oxide Capacitance of the MOS Transistor in (F) Farads Cgs- gate to Source Capacitance of the MOS Transistor in (F) Farads Avt – Slope Factor Of Threshold Voltage in (mVµm) σvt - Standard deviation of the Threshold Voltage in mV vi Contents 1 PMU And The Need for NMOS LDOs .........................1 1.1 Power Management Unit (PMU).........................................................................1 1.2 DC/DC Converters.............................................................................................2 1.3 Low Drop Out (LDO) Regulators .......................................................................2 1.4 Introduction to NMOS LDOs ...............................................................................3 1.5 PMOS Low Drop Out Regulator..........................................................................3 1.6 Low Voltage Drive ...............................................................................................4 1.7 Specifications of the NMOS Regulators..............................................................5 1.8 Summary ..............................................................................................................7 1.9 References ............................................................................................................7 2 PMOS Low Drop Out Regulators ...............................9 2.1 Choice of the Dimension of the PMOS Power Transistor ...................................9 2.2 Conventional Design Approach for the PMOS LDO...........................................9 2.3 Load Capacitor with a certain ESR and ESL ....................................................11 2.4 ESR Zero ............................................................................................................12 2.5 Compensation Scheme for PMOS LDO.............................................................13 2.6 Summary ............................................................................................................13 2.7 References ..........................................................................................................13 3 Conventional Design Of NMOS LDOs .......................15 3.1 Architecture of the NMOS LDO.........................................................................15 3.2 Dimensioning The NMOS Power Transistor .....................................................15 3.3 Stability of the NMOS LDO ...............................................................................17 3.4 Conventional Design Approach-I ......................................................................19 3.5 Conventional Design Approach -II....................................................................19 3.6 Pole-Zero Calculations For Conventional-I Approach.....................................20 3.7 Pole-Zero Calculations For Conventional Design Approach-II ......................22 3.8 Implementation Of The Conventional Circuit...................................................24 3.9 AC and Transient Response Of The NMOS LDO .............................................26 3.10 Problems With The Scalability Requirement ...................................................29 3.11 Summary ..........................................................................................................31 3.12 References ........................................................................................................31 vii 4 Compensation Of The NMOS LDO.........................33 4.1 Introduction.......................................................................................................33 4.2 Proposed Compensation with Small Signal Model............................................33 4.3 Small Signal Analysis.........................................................................................34 4.4 Pole Zero Locations for Full Load and No Load Conditions...........................35 4.5 Choice of the Compensation Capacitor Cc ........................................................36 4.7 Trade Off At No Load and Small Load Currents..............................................39 4.8 Implementation Of The Miller Compensated Circuit .......................................40 4.9 AC Response And Transient Characteristics....................................................42 4.10 Ringing in Time Domain and Poor Gain Margin in AC Domain....................45 4.11 High Frequency Peaking ................................................................................46 4.12 Summary .........................................................................................................46 4.13 References .......................................................................................................46 5 Damping Factor Enhancement, Power Sense Construction with Simulation Results............................47 5.1 Introduction.......................................................................................................47 5.2 Non-Dominant Poles.........................................................................................47 5.3 Improved Damping ............................................................................................49 5.4 Power Sense Construction .................................................................................50 5.5 Peaking At High Frequencies ............................................................................50 5.6 AC Characteristics.............................................................................................53 5.7 Transient Response ............................................................................................54 5.8 DC Response......................................................................................................54 5.9 PSRR Calculations.............................................................................................56 5.10 Monte-Carlo Simulations.................................................................................56 5.11 Equivalent Output Noise ..................................................................................57 5.12 Summary ..........................................................................................................58 5.13 References ........................................................................................................58 6 Additional Features In The NMOS LDO..................59 6.1 Programmability of the output voltage ..............................................................59 6.2 ECO Mode .........................................................................................................61 6.3 Adaptability Of The LDO To Scalable Load Currents ......................................64 6.4 Cap Free Option ................................................................................................66 viii 6.5 Summary ............................................................................................................66 7 Conclusions And Future Work..................................67 Future Work .............................................................................................................67 Appendix- I...............................................................................................................68 Appendix- II .............................................................................................................73 1 1 PMU And The Need for NMOS LDOs ____________________________________________________________________ We live in an era where we like to access information when we are on the move. Electronic gadgets like the cellular phones, mobile audio, video systems like the Apple IPODs become more complex. More power is consumed by both active and standby systems. Consequently, power-management design for portable wireless devices imposes new challenges in the areas of I/O interface, energy management and battery lifetime. Power regulation and management IC’s have become one of the fastest growing segments of the electronics industry largely due to the proliferation of portable electronic devices such as cell-phones, MP3 players, PDA’s, and game machines. This has led to increasing demand for higher levels of integration in order to reduce boardspace requirements and lower the BOM (Bill of materials) [2]. 1.1 Power Management Unit (PMU) Battery is the only source of power that we can use to supply the circuits. Battery management is very important in this mobile era. We need to ensure that optimum use is made of the energy inside the battery that powers the portable product [1]. Power management involves the implementation of functions that ensure a proper distribution of power throughout the system ensuring minimum power consumption by each system. Examples are active hardware and software design changes for minimizing power consumption, such as reducing clock rates in digital system parts and powering down system parts that are not in use. A PMU consists of DC-DC converters [1] and LDOs (Low Drop Out Regulators) [1]. An overview of a PMU is given in the Figure (1.1). LDO DIGITAL Battery LDO ANALOG DC/DC CONVERTER Load Currents Figure (1.1) A Typical Power Management Unit 2 1.2 DC/DC Converters DC/DC converters convert one DC voltage to another DC voltage. The main input DC voltage is the Lithium ion battery voltage that always stays between 2.75V to 4.25V. The generated output voltage may need to be even higher than the input DC voltages as in the display driver applications [1]. Apart from that, the digital circuits operate at voltages lower than 1V. This conversion is done by the DC-DC converters. In general, two approaches exist for converting one voltage into another. The first approach involves a time-continuous circuit with a dissipative element, whereas the second approach involves a time-discrete circuit with an energystorage element. The first approach is only suitable for converting a higher voltage into a lower voltage, which is down-conversion, whereas the second approach enables upand down-conversion. The two approaches are illustrated in the Figure (1.2) in which the battery voltage Vin is converted into Vout. Figure(1.2) Two approaches to convert a battery voltage into another voltage. (a): Time continuous with dissipative element (Vin > Vout only) (b) Time-discrete with energy-storage element The dissipative element in the Figure (1.2a) remains connected between the battery and the load. The efficiency of the voltage conversion will always be lower than 100%, because of its dissipative nature. The energy-storage element in Figure (1.2b) is first connected to the battery to store energy, after which it is connected to the load to supply this energy. The efficiency of the voltage conversion process is 100% in the theoretical case in which no energy is lost in the energy-storage element and switches. The type of time-discrete voltage converter employed and the characteristics of the employed components will determine the efficiency in practice. An energy buffer Cbuf is necessary, because of the time-discrete nature of converters of this type. 1.3 Low Drop Out (LDO) Regulators Regulate as the name clearly says, the function is to regulate the output voltage despite the variations in the load currents. In other words a LDO is a voltage source which delivers a constant output voltage despite variations in the load current. The low noise characteristics and smaller size and complexity of these regulators, make them ideal candidates for many applications, especially where the (Vin-Vout) difference is small. 3 For very low power applications, linear regulators are actually preferable even if the efficiency is low not only due to the lower cost and complexity, but also because their quiescent currents are lower. Due to their lower noise contribution, LDO regulators are favoured for powering many sections of the typical cellular handset. LDO’s are very suitable for powering the baseband, RF, TCXO, RTC, and audio sections of the typical handset. They are also being used to power white LED’s for backlighting in other portable applications. Each of these applications has different requirements which have traditionally been met by off-chip regulator IC’s. Keeping the regulators off-chip not only increases BOM cost, it also lowers system reliability, requires valuable board space and creates more stringent requirements on the regulator due to losses on the PC-board itself. 1.4 Introduction to NMOS LDOs To overcome the disadvantage mentioned in Section 1.3, in modern PMUs the DC-DC converters are used to convert one voltage level to the another because of their high efficiency. The DC-DC converters have ripple on their outputs. In order to remove the ripple, LDOs are used as the post regulators to regulate the output voltage of the DC-DC converters. This is pictorially represented in the Figure (1.3). DC/DC CONVERTER LDO LDO Battery DC/DC CONVERTER LDO Load Current Figure(1.3) LDOs as Post Regulators 1.5 PMOS Low Drop Out Regulator Most of the existing LDOs are PMOS LDOs. It is discussed here in order to give an overview of the design approaches for a LDO. A schematic representation of a PMOS LDO is shown in Figure (1.4). 4 It consists of a PMOS transistor, controlled by an error amplifier, which compares a fraction of Vldo_out with a reference voltage Vref. The transistor (MP_LDO) is operated in the saturation region [3]. The error amplifier is supplied by the VISA ( Voltage Internal Supply Analog). (Vldo_in - Vldo_out), which is the dropout voltage, has to be present for proper operation. MP_LDO PMOS, Power Transistor Vref Vldo_out VISA - Vgate + VSS R1 R2 Vldo in Vgate Vldo out Figure (1.4) PMOS LDO Regulator 1.6 Low Voltage Drive As we move towards low voltage ranges the input to the LDOs from the DCDC converters will be getting lower and lower. When we think of the option of a PMOS LDO, we can easily see that the source voltage of the PMOS power transistor in the Figure (1.4) is getting lesser than VISA which also needs a low gate voltage even lesser than VSS in order to drive a certain load. The error amplifier will no more function in this case and we need DC-DC converters that generate voltages lower than VSS in order to supply the error amplifier and to pull down the gate voltage below the VSS level. The other solution is to replace 5 the PMOS power transistor in the Figure (1.4) with an NMOS Power transistor, paving way to the design of the NMOS LDO regulators. This is shown in the Figure(1.5). Vldo_in Vref + MN_LDO NMOS, Power Transistor Vldo_out VISA Vgate - VSS R1 R2 Vldo in Vgate Vldo out Figure (1.5) NMOS LDO Regulator As shown in the Figure (1.5) even if the Vldo_in level gets lower, the gate voltage of the NMOS power transistor needs to stay at a positive level (i.e. a positive Vgs) in order to drive the NMOS transistor for a certain load current. Like the PMOS LDO regulator, we need to bias the NMOS Power transistor in the NMOS LDO in the saturation region [3]. Hence we see a need for NMOS LDO regulators when the regulator inputs get lower. 1.7 Specifications of the NMOS Regulators The specifications of the NMOS LDO Regulators are shown in Table(1.1).These specifications have to be translated in terms of the design parameters like the open loop gain of the LDO, UGB etc. A higher open loop gain will result in lower offset voltage [3] which translates into the accuracy specification of the output voltage. Load step indicates a measure of the stability and speed of the 6 LDO. The UGB is a measure in terms of speed of the LDO translating to the overshoot and undershoot during the load step. The reference voltage has to be chosen based on the regulator input and the output voltage to be delivered. The drop out voltage is 300mV. During stand by mode of the application, the LDO needs to draw very less current and should be stable. Table 1.1 Detailed Specifications Of The NMOS LDO Tamb: -40°C to 85°C Output current within nominal range; unless otherwise specified. Symbol Parameter Conditions Min Typ Max Unit Vldo_in 1.8 V 1.5 V V mV Vldo_out Vs Regulator supply voltage 0.9 Programmable output voltage Programming step size 0.6 (Vp+Vdropout) 25 Vpacc Absolute accuracy of output voltage Iout=10% of Ioutmax 1 % Vpload Load regulation delta Iout 1mA....Ioutmax 1 % Vpline VISA Line regulation Internal analog supply voltage Dropout voltage External reference voltage output current in normal mode output current in eco mode in % of Ioutnormal Regulator Output Capacitor Series resistance load capacitor Total supply current in active mode Total supply current in ECO mode Total supply current in OFF mode relative to Vin band-gap off band-gap on Iout=Ioutmax 0.1 2.7 2.45 300 0.605 %/V V mV V 500 mA 10 % 470 nF Vdropout Vref Ioutnormal Iouteco Cdec ESR_Cdec Isupact Isupeco Isupoff Scalable ESL<2nH Iout=0mA 2.1 2.35 2.4 2.4 0.595 0.600 50 100 5 mΩ As low as possible (50µ +0.25%*Iout) As low as possible (15µ +1%*Iout) 1 µA *Note: Vdrop is defined as the Vldo_in – Vldo_out voltage, when Vldo_out is 1% below the set output voltage. The voltage drop over the bond wires is included 7 1.8 Summary In this chapter, the PMU and the LDOs were introduced. The need for the NMOS LDO has been discussed with its specifications. 1.9 References [1] Battery Management Systems, A Design by Modelling by H.J.Bergveld, Ph.D Thesis , Universiteit Twente, June 2001. [2] Power Management Unit ,Electronic Letters [3] Analog Integrated Circuit Design by B.Razavi. 8 9 2 PMOS Low Drop Out Regulators _____________________________________________________________________ 2.1 Choice of the Dimension of the PMOS Power Transistor In modern PMUs the DC-DC converters are used in order to convert one DC level to another. The DC-DC converters have high efficiency. But the output voltage in DC-DC converters has a lot of ripple [1]. The main reason why a LDO would be used after the DC-DC converter is to remove this ripple. Hence the Power Supply Ripple Rejection (PSRR)[1] is a very important item on the list of boundary conditions. In Section 1.3 we refer LDO to be a voltage source which means that the LDO should have a low output impedance. The power transistor is dimensioned based on its current and |Vgs-Vth|. Calculations in [3] show that a transistor with 1V overdrive may need thousands of micron wide transistor to handle hundreds of milliamps of currents. Also a large W/L is opted in order to maintain a low drop out voltage. A larger W/L helps to dissipate the heat within its area whereas a smaller W/L spreads the heat to the near by transistors, thereby causing problems. On the other hand the total dimension of the MP_LDO has to be minimized because of chip costs and gate current (due to parasitic gate capacitors). A trade-off has to be made at this juncture. 2.2 Conventional Design Approach for the PMOS LDO From the Figure (1.4) we can see that the open loop gain of the PMOS LDO is the product of the gains of the error amplifier and the gain of the single stage amplifier built by using the MP_LDO transistor. This means that we can see a two pole [2] transfer function, one pole coming from the output of the amplifier and the other from the MP_LDO and the load capacitor. The transfer function can be given as AoL ( s ) = Aerror _ amp * AMP _ LDO ( s + ω pa )( s + ω pL) → (2.1) where AoL(s) is the open loop gain of the PMOS LDO Aerror_amp is the gain of the error amplifier AMP_LDO is the gain of the output stage of the LDO with the Power Transistor ωpa is the pole from the amplifier ωpL is the pole from the Output Stage For a stable system, the poles have to be far apart with the second pole far off from the Unity Gain BandWidth (UGB)[2] of the LDO. Unity Gain Bandwidth Will be determined by the capacitor that sets the dominant pole along with the transconductance of the error amplifier that sets the gain.[3] 10 Line and load transient measurements show the LDO’s ability to respond to abrupt changes in line voltage and load currents. These test reveal significant overshoot, or sustained ringing in the output as it attempts to maintain regulation. We need to have a response without any overshoots and ringing at the output. The basic step for this is to place a capacitor at the output which will remove the overshoots. At the same time the ringing of the LDO shouldn’t occur by placing this capacitor. This ringing is a measure of the phase margin of the LDO when we talk about its gain and phase characteristics in open loop. Equation (2.1) can be explained a bit more in detail. The gain of the amplifier stages are specified in terms of a certain transcondutance and output impedance. If the transconductances of the error amplifier and the output PMOS power transistor MP_LDO are gma ,gmo respectively with their output impedances Ra and Ro respectively then the overall open loop gain of the LDO can be expressed as AoL( s ) = gmaRagmoRo ( s + RaCa )( s + RoCL) → (2.2) Ca and CL are the capacitance seen at the output of the amplifier and the load capacitance respectively. It is a very common design approach to have the LDO, output pole dominant [1]. This means that the dominant pole is determined by the load capacitor. The expression for the dominant pole fpL will be fpL = 1 2π RoCL → (2.3) The UGB can be expressed as fτ . fτ = gmaRagmo 2π CL → (2.4) Now the design technique is to keep the pole from the error amplifier outside the UGB. To do that we need a low output impedance from the amplifier. By sacrificing the gain of the error amplifier, we achieve low output impedance and thereby the stability is obtained. 11 2.3 Load Capacitor with a certain ESR and ESL We connect an external capacitor in order to improve the transient response of the LDO. This is a Ceramic SMD (Surface Mounted Device) Capacitor. This external capacitor is not ideally capacitive in nature [13]. The capacitor has with it an Equivalent Series Resistance (ESR) and an Equivalent Series Inductance (ESL). Hence the load seen by the LDO is an R-L-C circuit rather than a purely capacitive one. The realistic capacitor is shown in the Figure (2.2). Figure (2.2) Realistic Ceramic SMD Capacitor Apart from that we also have the inductance of the bond wire from the bond pad to the pin and the routing tracks from the pin to the load capacitor offer some inductance and resistance. Hence the realistic load the LDO sees is shown in Figure (2.3). This is used in the simulations. The capacitor should be placed as close as possible to the pin because placing it a bit far away from the pin results in a larger track inductance of the order of 10nH, which could result in stability problems [2].A bondwire is made out of Gold metal. It has a diameter of 25µm. The resistance of the bondwire for a 1mm length can be calculated as 12 R= ρL A ' ρ ' is the specific resistance of the Gold wire of the order of 2.2e − 8 Ωm @ Temperature T = 293K ⇒R= 2.2e − 8*1e − 3 ~ 50mΩ π (12.5e − 6) 2 In most of the situations the bondwire is 1mm to 2mm long and so the bondwire resistance is of the order of 50mΩ to 100mΩ. Also the bondwire of 1mm length has an inductance of 1nH. To lower the bondwire resistance ,two bondwires are used in parallel offering a resistance of 25mm and an inductance of 0.5nH. Bond Wire 50mΩ 1nH LDO Bond Pad Track Resistance 5mΩ Track Inductance 2nH Load Current ESR = 5mΩ ESL = 2nH Capacitor = 470nF Figure (2.3) Realistic Load of the LDO 2.4 ESR Zero The ESR along with the load capacitor produces a zero. A higher ESR pushes the zero towards low frequencies. This would be a Left Half Plane Zero [4] which produces a positive phase shift. At higher frequencies the inductive reactance increases and a RLC combination could produce complex poles and 13 zeroes which will have a big impact on the stability of the LDO and are discussed in the ensuing chapters. 2.5 Compensation Scheme for PMOS LDO An output pole compensated PMOS LDO will often result in the choice of a big capacitor when large load currents are handled. This big capacitor helps the output pole to be restricted within a certain KHz band. Then at the expense of a reasonable current it may be easy to place the poles of the amplifier beyond the UGB. Also in many designs, the amplifier pole is pushed to a low frequency region and is cancelled by the zero from the ESR [4]. Pushing the amplifier to the low frequency region gains us the gain of the LDO. Unfortunately these approaches demand the need of a big output capacitor. A smaller capacitor takes a smaller area and so the integration on the Printed Circuit Board (PCB) becomes easier. Frequency compensation techniques have been developed in order to address the stability of the LDO with a smaller load capacitor. The compensation techniques deal with pole-splitting techniques widely discussed and verified in [5,6,7,8,9,10]. In these compensation techniques the dominant pole changes to the amplifier pole while the load pole is moved beyond the UGB or is cancelled with an ESR Zero. The ESR of these capacitors is of the order of 200mΩ. In [11] another approach of mirroring the load current is suggested where the variation in the load pole is compensated by a varying zero. In [12] another compensation technique has been proposed. This compensation scheme introduces a zero that is independent of the ESR of the capacitor and this would work only when this ESR is low, possibly lesser than 50mΩ. 2.6 Summary In short an overview of the functionality of the PMOS LDO and the techniques to get them stable have been discussed. 2.7 References [1] LDO Topologies by G.E.de Vrieze, March 29th,2000, Philips Semiconductors, Internal Document. [2] Line and Load Transient Testing for Power Supplies, Dallas Semiconductor,Maxim Editorial on Power Supply Circuits,Dec30,2004. [3] Analog Integrated Circuit Design by B.Razavi [4] Technical Review Of Low Drop Out Voltage Regulator, Operation and Performance, Application Report, Texas Instruments Ltd. [5] Operational Amplifiers, Theory and Design by Johan.H.Huising, Kluwer Academic Publishers. [6] G. A. Rincon-Mora and P. E. Allen, “A low-voltage, low quiescent Current low drop-out regulator,” IEEE J. Solid-State Circuits,vol. 33, pp.36–44, Jan. 1998. [7] “A new method for multiplying the Miller capacitance using active components ,” by G.de Cremoux, Y.Christoforou, I.van Loo [8] “Optimized frequency-shaping circuit topologies for LDOs,” IEEE Trans. Circuit and Systems II, vol. 45, pp. 703–708, June 1998. [9] G. A. Rincon-Mora, “Active Capacitor Multiplier in Miller Compensated Circuits,”, IEEE J. Solid-State Circuits, vol.35, pp.26-32,Jan.2000. 14 [10] Ka Nang Leung and Philip K.T.Mok , “A Capacitor-Free CMOS Low Dropout Regulator With Damping Factor Control Frequency Compensation,”, IEEE J. Solid-State Circuits, vol.38, pp.16911702,Oct.2003. [11] Ka Chun Kwok and Philip K.T. Mok , “Pole Zero Tracking Frequency Compensation For Low Dropout Regulator, ” [12] “A Frequency Compensation Scheme for LDO Voltage Regulators,” by K.Chava and J.Silva-Martinez, IEEE J. Solid-State Circuits, vol.51, June 2004 [13] “ Improve Your Designs with Large Capacitance Value Multi-Layer Ceramic Chip ( MLCC ) Capacitors ” by George M. Harayda, Akira Omi, & Axel Yamamoto, from Panasonic Electronic Equipments. 15 3 Conventional Design Of NMOS LDOs _____________________________________________________________________ 3.1 Architecture of the NMOS LDO The architecture of the NMOS LDO is shown in the Figure (3.1). The NMOS LDO consists of an error amplifier and a level shifter. The error amplifier of the LDO is supplied by an Internal Analog Supply Voltage(VISA) and the battery voltage VBAT is used to supply the Level Shifter. The level shifter is used in order to boost the gate voltage of the power transistor to a level greater than the VISA level. This is essential to handle large load currents with an optimal dimension of the power transistor. A bigger power transistor results in large parasitic capacitances there by hampering the speed of the LDO. The Input Voltage to the LDO (Vldo_in) ranges between 1.8V to 0.9V. The output voltage (Vldo_out) is between 1.5V to 0.6V. The load current is from 0 to 500mA at a quiescent current consumption of 50µA with a load regulation accuracy of 1%. Vbat Vldo_in (Regulator Input) Level shifter Error Amplifier Vref + - Power Transistor Vgate MN_LDO VISA Vldo_out (Regulator Output) VSS R1 R2 IL: Load Current = 500mA Figure (3.1) Proposed NMOS LDO Architecture 3.2 Dimensioning The NMOS Power Transistor The first step in the design is to dimension the power transistor. The transistor should be biased in the saturation region [1].The maximum current through the transistor and the drop out voltage are known. The process parameters like the threshold voltage (Vth), process gain µnCox of the transistor are known. For our output requirements, we know that we need a low voltage NMOS transistor. It is available in the Philips C050PMU process [1]. The normal NMOS transistor suffers from body effect [2] which in turn will increase the threshold voltage of the transistor thereby demanding an increase in the dimension of the transistor or demanding a higher over drive to handle a certain current . 16 Hence in Philips C050PMU process [1], special NMOS transistor called as the Isolated NMOS transistor is available, whose body effect is zero because the source and the bulk of the NMOS transistor can be tied together. The Figure (3.2) shows the cross section and the symbol of the NMOSISO transistor. The PW1 (special Pwell) which is the bulk of the transistor is tied to the source. The Nwell is tied to the drain. ProMOST, the Internal Philips tool helps us to determine the W/L of the transistor, for a certain Vds, Vgs-Vth and drain current Id. Table 3.1 gives us an insight of the needed W/L for a certain Vgs-Vth. The conclusion is that if we can boost the gate of the NMOS Power Transistor to 2.5V even if the VISA is 2.2V, we would be able to optimize its dimension .From[1] we infer that the maximum gate voltage that can be applied to the gate of the NMOS power transistor is 2.5V. So now we choose between our voltage that we need to generate and the W/L. n+ n + Nwell Gate PW1 PW1 NWELL Substrate Figure (3.2) NMOSISO transistor ; Cross Section and Symbol Table 3.1 Necessary (W/L) of The Power Transistor (vs) its Vgs, Vds, Id S.No 1. 2. 3. 4. 6. 7. Current through the NMOS Power Transistor 500mA 500mA 500mA 500mA 500mA 500mA W/L of the NMOS Power Transistor 375,000µ/0.25µ 50,000µ/0.25µ 40,000µ/0.25µ 30,000µ/0.25µ 20,000µ/0.25µ 11,500µ/0.25µ Gate Voltage of the NMOS Power Transistor 2.16V 2.28V 2.30V 2.33V 2.40 2.5V Vldo_in (Drain Voltage) Vldo_out (Source Voltage) 1.8 1.8 1.8 1.8 1.8 1.8 1.49994 1.50003 1.50001 1.49998 1.49999 1.4996 17 3.3 Stability of the NMOS LDO We have to map the Figure (3.1) into the small signal model which is shown in, the Figure (3.3) from which we analyze the stability. The stability of the NMOS LDO is analyzed by the Bode plot that is obtained from the Open Loop Simulation of the LDO. Vbat Vldo_in (Regulator Input) Level shifter Error Amplifier Vgate MN_LDO Vref + - VISA Vldo_out (Regulator Output) VSS Ra Loop opened here. This path shown in dotted lines is open for AC and closed for DC. Rb V1 Vref gmxVx βVo gm1V1 gmL(VL-Vy) gmp(VL-Vo) Figure (3.3) Large Signal and Small Signal Model of the NMOS LDO Error Amplifier is modeled as a two stage amplifier. First stage is a differential stage without any high impedance node to obtain a voltage gain. Hence it is modeled as a subtracting node that subtracts (Vref-βVo) where ‘β’ is the Rb to 1 when the output voltage is feedback fraction. It varies from Ra + Rb programmable from 1.5V to 0.6V. 18 Output Impedance of the first stage is R1||C1. Output impedance of the output stage is Rx||Cx. Level Shifter is modeled as a Voltage Controlled Current Source (VCCS) with an output impedance, RL = 1 gmL where gmL is the transconductance of the LevelShifter . Power transistor is modeled as a VCCS with an output impedance equal to the transconductance of the Power Transistor. Cp is the Gate to Source Capacitance of the Power transistor. Load Capacitor has an Equivalent Series Resistance (ESR) and Equivalent Series Inductance (ESL) associated with it. The ESL is omitted to make the mathematical expressions that are calculated from the small signal model simpler. g mp ; Load Pole; 2π C L g mp is the transconductance of the power transistor ; f p1 = C L is the load capacitor fpa 1 ; Amplifier Pole; R x Output impedance of the amplifier . 2π R xC x C x is the Capacitance seen at the output of the amplifier = Pole due to R 1C 1 can be neglected if we design such that R 1C 1 << R xC x Load capacitor sets the dominant pole and LDO gain rolls off after fp1 As a result , the input impedance of the power transistor is its C gs and so there occurs a pole due to the levelshifter at fpL = g mL 2π Cp ( output impedance of the levelshifter = fz = 1 ) g mL 1 ; Zero due to the ESR & the Load Cap 2π R esrC L The pole from the power transistor moves from low frequencies in Hertz to mid frequencies in KHz. For a particular load current, the pole from the power transistor may clash with the pole from the error amplifier which leaves the LDO unstable. 19 3.4 Conventional Design Approach-I In the conventional designs, the LDO has a dominant pole from the power transistor. The remaining poles are moved outside the Unity Gain Bandwidth (UGB). This has a big disadvantage of burning a lot of quiescent current to reduce the output impedance of the error amplifier in order to move the error amplifier pole outside the UGB, which in fact has to result in a lower output impedance. Lower output impedance lowers the DC gain which in turn affects our output accuracy. The Bode plot is shown in the Figure (3.4). fp,no load fp,full load x x Gain (dB) Gain and phase for No Load Condition 0 Phase (degree) x amplifier, fz,esr fp,error 0 -45 ° -90 ° -180° Frequency Figure (3.4) Bode Plot of a Conventional Design Approach in a NMOS LDO 3.5 Conventional Design Approach -II If we target a moderate gain of the LDO we increase the output impedance of the error amplifier and move pole of the amplifier within the UGB. Choosing a big load capacitor along with the ESR and bond wire resistance could produce a zero within the UGB. This would cancel the pole from the error amplifier and leaves us with an output pole dominant LDO. This is shown in Figure (3.5). We place the pole of the level shifter outside the UGB and so it is not shown in the Figure(3.5). 20 fp,no load x Gain (dB) fp,full load x x fp,error amplifier, fz,esr 0 Phase (degree) 0 -45° -90° -180° Frequency Figure (3.5) Bode Plot of a Conventional Design Approach in a NMOS LDO 3.6 Pole-Zero Calculations For Conventional-I Approach The position of the poles and zeroes can be calculated as follows. Let us assume the following values for full load which is 500 mA. Conventional Approach − I g mp = 1.5 S ; R x = 10 M Ω ; C x = 20 fF ; g mL = 400 e − 6 S ; C p = 6.65 pF g m1 = 20 e − 6 S ; R 1 = 48 K Ω ; C 1 = 10 fF ; g mx = 20 e − 6 S . C L = 470 nF ; R esr = 30 mΩ (including the bondwire resistance ) DCGain = g m1 R1 g mxR x = 20 e − 6 * 48e + 3 * 20 e − 6 *10 e + 6 = 192 = 45 dB 1 1 fpa = ~ 800 KHz = 2π R xC x 2 * 3.141*10 e + 6 * 20 e − 15 g mp 1.5 fp1 = ~ 510 KHz = 2π C L 2 * 3.141* 470 e − 9 g mL 400 e − 6 fpL = ~ 10 MHz = 2π C p 2 * 3.141* 6.65e − 12 1 1 ~ 11MHz fz 1 = = 2π R e srC L 2 * 3.141* 30 e − 3 * 470 e − 9 The UGB is given by fτ = 1 2π g m1 R 1 g mxg mp 1 = C LC x 2π 20 e − 6 * 48e + 3 * 20 e − 6 *1.5 ~ 8.8 M Hz 470 e − 9 * 20 e − 15 Under the full load condition, the LDO is unstable as there are two poles within the UGB. 21 Gain 45 (dB) fp1 fpa fτ 0 fpL, fz1 Frequency Phase (degree) 0° -45° -90° -135° -180° Frequency Figure (3.5a) Conventional Approach- I : Location of Poles and Zeroes We can see from the calculations and its Bode-Plot in the Figure (3.5a) that the phase margin is poor because of the two poles lying with in the UGB. This reflects the ringing behavior during the load step in the transient. In order to get the LDO stable with a good phase Margin of 60° as per the conventional approach-I, we need to push the pole of the error amplifier outside the UGB. This means that we should reduce the value Rx by a factor of 12 so that we move the pole of the error amplifier outside the UGB leaving the LDO output pole dominant. Hence when we decrease the output impedance of the error amplifier by a factor of 12, which in turn decreases the gain to 24dB which is less for accuracy. Under no load condition, gmp would decrease to say 347µS which moves the load pole to 117Hz and would also decrease the UGB proportionately. But the error amplifier pole is not that far away from the UGB thereby affecting the stability. The conclusion is that the LDO is stable for currents only if the error amplifier pole is moved beyond the UGB at the expense of the decrease in the gain. This is explained graphically in the Figure (3.5a). 22 3.7 Pole-Zero Calculations For Conventional Design Approach-II By choosing a little bigger load capacitor, let us calculate our poles and Zeroes. Let us assume the following values for full load which is 500 mA. Conventional Approach − II gmp = 1.5 S ; Rx = 10 M Ω; C x = 20 fF ; gmL = 400e − 6 S ; C p = 6.65 pF gm1 = 20e − 6 S ; R1 = 48 K Ω; C 1 = 10 fF ; gmx = 20e − 6 S . C L = 4.7 µ F ; Resr = 30 mΩ (including the bondwire resistance ) DC Gain = gm1 R1 gmxR x = 20e − 6 * 48e + 3* 20e − 6 *10e + 6 = 192 = 45dB 1 1 ~ 800 KHz = 2π R xC x 2 * 3.141*10e + 6 * 20e − 15 gmp 1.5 f p1 = ~ 51KHz = 2π C L 2 * 3.141* 4.7 e − 6 gmL 400e − 6 fpL = ~ 10 MHz = 2π C p 2 * 3.141* 6.65e − 12 1 1 fz1 = = ~ 1.1MHz 2π ResrC L 2 * 3.141* 30e − 3* 4.7 e − 6 fpa = The UGB is given by fτ = 1 ⎛ gm1 R1 gmxgmpResr ⎞ ⎜ ⎟ 2π ⎝ Cx ⎠ fτ = 1 ⎛ 20e − 6 * 48e + 3* 20e − 6 *1.5 * 35e − 3 ⎞ ⎜ ⎟ ~ 8 MH z 2π ⎝ 20e − 15 ⎠ Under the full load condition, the LDO is stable as there are two poles and a zero within the UGB. Here we can see that the error amplifier pole and the ESR zero lie within the UGB. As a result of this, the -90° phase shift by the error amplifier pole would be compensated by a +90° phase shift from the ESR zero there by providing a good phase margin. This is shown graphically in the Figure (3.5b). 23 Gain 45 (dB) fp1 fpa, fz1 fτ 0 Phase (degree) fpL Frequency 0° -45° -90° -135° -180° Frequency Figure (3.5b) Conventional Approach- II ; Location of Poles and Zeroes From the Figure (3.5b) we see that the phase margin is 49˚. Phase Margin ' Φ' is calculated by using the relation : ⎛ fτ ⎞ ⎛ fτ ⎞ ⎛ fτ ⎞ ⎛ fτ ⎞ Φ = 180°− tan−1 ⎜ ⎟ − tan−1 ⎜ ⎟ + tan−1 ⎜ ⎟ − tan−1 ⎜ ⎟ ⎝ fp1 ⎠ ⎝ fpa ⎠ ⎝ fz1 ⎠ ⎝ fp2 ⎠ ⎛ 8e + 6 ⎞ −1 ⎛ 8e + 6 ⎞ −1 ⎛ 8e + 6 ⎞ −1 ⎛ 8e+ 6 ⎞ ⇒Φ = 180°− tan−1 ⎜ ⎟ ⎟ − tan ⎜ ⎟ + tan ⎜ ⎟ − tan ⎜ ⎝ 10e + 6 ⎠ ⎝ 51e + 3 ⎠ ⎝ 800e + 3 ⎠ ⎝ 1.1e + 6 ⎠ ⇒Φ = 180°− 90°− 84°+ 82°− 39° ⇒Φ = 49° 24 3.8 Implementation Of The Conventional Circuit The circuit that is realized from the small signal model is shown in the Figure (3.6). The transistors MP135 and MP136 form the differential pair biased by a current source of 15µA. The reference voltage is supplied to the non-inverting terminal of the differential amplifier. The currents from the transistors MP135 and MP136 are mirrored through the NMOS current mirrors MN1, MN3 and MN2, MN4. Additional current sources have been added into MN1 and MN2 in order to control the output impedance of the error amplifier independent of the gm of the differential stage. The current in MN3 is mirrored again by the PMOS current mirror MP134, MP138. The drain of MP138 and MN4 is a high impedance point where we obtain the voltage gain of the error amplifier. The output voltage of the error amplifier drives the level shifter. The Level shifter is a simple implementation with a current source and a PMOS transistor. This provides an output voltage greater than the minimum supply (VISA) of the error amplifier. The output of the level shifter drives the NMOS power transistor. The dimension of the power transistor is chosen from the Table (3.1). It is opted to be 11500µ/0.25µ in order to handle a load current of 500mA. The source of the NMOS power transistor is connected to the resistive divider whose output voltage from a certain tap point is fed back to the inverting terminal of the differential amplifier. The source of the NMOS LDO which is our output point is connected to the pin via the bond pad and the bond wires. The Load capacitor is connected to the pin. The load capacitor has a certain ESR and ESL of around 5mΩ and 2nH. The bond wires (two bondwires in parallel) have a resistance of 25mΩ to 30mΩ to and an inductance of 0.5nH (i.e. 1nH/mm). The resistance produces a zero with in the UGB which helps to improve the phase margin. This has a disadvantage that we have voltage drop of Iload*30 mΩ. For larger load currents of the order of 500mA, the voltage drop would be around 15mV. When the output targeted is 1.5V, we would achieve 1.485V after the 15mV drop which reflects that an accuracy of 99% or the error in the output voltage is 1%. %error = 1.5 − 1.485 *100 = 1% 1.5 When our regulator input would be only 400mV, then we would have a voltage drop of 15mV, leading to an output voltage of 385mV. The error is 3.75% and is not acceptable. %error = 0.4 − 0.385 *100 = 3.75% 0.4 So this problem can be overcome and is explained in the later chapters. The AC response and the transient characteristics are discussed in section 3.8. 25 Figure (3.6) Realized Transistor Topology for NMOS LDO 26 3.9 AC and Transient Response Of The NMOS LDO The open loop simulation set up is shown below in Figure (3.7). The loop is opened at the source of the NMOS LDO. This means that the resistor divider is disconnected from the source of the NMOSLDO for an AC signal. However it is closed for the DC levels. This is implemented by using the switches that are available in the Library “analogLib”. Two switches are used with one switch closed (“OFF” state) for DC and open (“ON” state) for AC, through which an AC signal of amplitude 1V is applied. The other switch is open for DC and closed for AC. Vbat (min)=2.75V, Vdd(min)=2.2V(VISA); iload =0 till 500mA; Load Capacitor Cdec=4.7µF Figure(3.7) Open Loop Simulation of the NMOS LDO The AC simulation results for a range of load currents are shown in the Figure (3.7) followed by the transient load step simulation results in Figure(3.8). Phase Margin is good for lower range load currents around 90° and as the load current increases the phase margin decreases as the load pole comes closer to the error amplifier pole. The ESR zero helps to stabilize the phase margin. 27 Load Current 0 to 20mA Load Current 50mA to 500mA Figure (3.8) AC response for load currents split in two graphs from 0 to 500mA 28 Figure (3.9) Transient Response For Rising and Falling Load Steps 29 From the AC characteristics we see that the phase margin is approximately 45° and in the transient response we see no oscillations for a load step of 0 to 500mA. 3.10 Problems With The Scalability Requirement As discussed in the specification of the NMOS LDO ( Section 1.7), we would like to have the LDO scalable. It means that, when the maximum load current would be set to 10 times a lower current value, then we would like to have our schematic in Figure (3.6) scaled down proportionately. For example, in the Schematic in Figure (3.6), the power transistor has a dimension of 11500µ/0.25µ. When the load current is scaled down by 10 times, then we would scale down our power transistor to 1150µ/0.25µ. At the same time it would really be nice to scale down the output capacitor by 10 times i.e. to 470nF from 4.7µF. We also would like to scale down the level shifter current and scale down the dimensions of all the transistors by a factor of 10. According to [3], the scaling down of all transistors by a factor of 10 will increase the offset of the LDO. This is because the W.L is reduced by a factor of 10 and the offset will increase by 10 as the offset is inversely proportional to the gate area according to the equation below, Avt WL Avt σ vt ( scaled ) = W L . 10 10 10 Avt ⇒ σ vt ( scaled ) = = 10* σ vt ( unscaled ) WL σ vt ( unscaled ) = Hence we don’t scale down the device dimensions. We scale down only our bias currents, dimension of the power transistor and the load capacitor. The bias current to the level shifter is not scaled down in order to retain the level shifter pole outside the UGB. After the load capacitor is scaled down, the ESR zero moves to the high frequencies because of the reduced load capacitor. This has a big impact on stability of the LDO because the ESR zero is moving to high frequencies and will not cancel the error amplifier pole as shown in the Figure (3.10) below. The calculations are shown below. 30 Let us assume the following values for full load which is 50mA. Conventional Approach − II for low load currents. gmp = 150e − 3S ; Rx = 10M Ω; Cx = 20 fF ; gmL = 400e − 6 S ; Cp = 665 fF gm1 = 20e − 6S ; R1 = 48K Ω; C1 = 10 fF ; gmx = 50e − 6S . CL = 470nF ; Re sr = 30mΩ (including the bondwire resistance) DCGain = gm1R1 gmxRx = 20e − 6* 48e + 3* 20e − 6*10e + 6 = 192 = 45dB 1 1 fpa = ~ 800 KHz = 2π RxCx 2*3.141*10e + 6* 20e − 15 gmp 150e − 3 fp1 = ~ 51KHz = 2π CL 2*3.141* 470e − 9 ⎛ ⎞ ⎜ ⎟ ⎜ ⎟ W ⎜ ⎟ 2 Id µCox gmp ⎜ ⎟ L ⎜ fp1 = fp 500 mA = 2π CL = ⎟ 2π CL ⎜ ⎟ ⎜ Since the current is 10 times less gmp scales down by 10 and 10 times smaller capacitor ⎟ ⎜ ⎟ W ⎜ ⎟ Id W 10 ⎜ ⎟ 2 µCox 2 Id µCox gmp ⎜ ⎟ 10 L = L = fp 500 mA ⎜ fp1 = 2π CL = ⎟ CL 2π CL 2*3.141* ⎜ ⎟ 10 ⎝ ⎠ So our load pole doesn ' t change. gmL 400e − 6 fpL = ~ 10MHz = 2π Cp 2*3.141*6.65e − 12 1 1 = fz1 = ~ 11MHz 2π Re srCL 2*3.141*30e − 3* 470e − 9 The UGB is given by 1 ⎛ gm1R1 gmxgmp R esr ⎞ 1 ⎛ 20e − 6* 48e + 3* 20e − 6*150e − 3*35e − 3 ⎞ ⎜ ⎟= ⎜ ⎟ ~ 8MHz Cx 2π ⎝ 20e − 15 ⎠ 2π ⎝ ⎠ Similarly the UGB doesn ' t change. Under the full load condition, the LDO is unstable as there are two poles within the UGB. fτ = 31 Gain 45 (dB) fp1 fpa fτ 0 fpL, fz1 Frequency Phase (degree) 0° -45° -90° -135° -180° Frequency Figure (3.10) AC Response Of the NMOS LDO showing the movement of Zero towards High Frequencies The conclusion is that the conventional Approach is not suited when scaling down the LDO requirements towards lower load currents with lower load capacitors. 3.11 Summary The conventional design approaches I and II for the NMOS LDO have been discussed. The stability issues have been discussed with small signal models , Bode Plots and equations. Also the load scalability and LDO adaptability have been discussed only to conclude that the conventional approach is not suited when we scale down the LDO requirements towards lower load currents with lower load capacitors. So frequency compensation is necessary in order to overcome the scalability and LDO adaptability issue. 3.12 References [1] Philips Semiconductors, C050PMU Process Manual. [2] Device Physics by Muller and Kamins. [3] Matching Properties of MOS Transistors by M.Pelgrom et al. 32 33 4 Compensation Of The NMOS LDO _____________________________________________________________________ 4.1 Introduction It is understood from Chapter 3, that we need an LDO which could be adapted when load currents scale down with the scaling down of the load capacitors. Hence we need a robust frequency compensation technique in order to design an LDO that meets the specification mentioned in the Section 1.7. Lot of frequency compensation techniques have been proposed for the operational amplifiers [1] and the PMOS LDOs [2-6]. A study has been made in order to apply a similar strategy for the NMOS LDO or if the topology in [7] can be adapted for the specifications in Section 3.2. The advantage of the topology in [7] is that ideally there is no pole from the power transistor as the power transistor is not a part of the feedback loop. Another way to design is to avoid the load capacitor, then the LDO can be made stable for all currents but this approach has a poor transient response. 4.2 Proposed Compensation with Small Signal Model The proposed compensation technique is shown in the Figure (4.1). The error amplifier has a differential input stage with current mirror loads. The compensation capacitor is connected across the input of the output stage of the error amplifier to the output of the NMOS LDO. This compensation is referred to as the Miller Compensation. Now the LDO has a dominant pole contributed by the Miller Capacitor and the output pole gmp due to the load capacitor is moved beyond , outside the UGB 2π CL guaranteeing a phase margin of 90°. In Section 4.3 we will also see that for the no load and small current ranges less than 1mA, the load pole is dominant and we don’t see the Miller Effect. The Phase Margin is only 25˚ but can be tolerated because of the short range of the currents. 34 V1 Vref gmxVx gm1V1 gmL(VL-Vy) gmp(VL-Vo) βVo Figure (4.1) Compensated NMOS LDO 4.3 Small Signal Analysis Writing the nodal equations at the nodes Vx, Vy ,VL and Vo to solve for we get ( See Appendix-I for the complete analysis) ⎡ sCc ⎤ − gmp ⎥ [1 + sCL Re sr ] gm1R1 gmxRx ⎢ Vo ⎣ gmxRx ⎦ = → (4.1) V 1 [− gmxgmpRxsCcR1 − {( gmp + sCL )(1 + sCxRx )(1 + sCcR1)}] The poles and zeroes are given by p1 = p2 = − g mp C L + (C xR x + g mxR xC cR1) gmp − ( C L + g mpC xR x + g mxg mpR xC cR 1 ) C LC xR x → (4.2) → (4.3) z1 = −1 R e srC L → (4.4) z2 = g mpg mxR x Cc → (4.5) Vo V1 35 Equation (4.1) shows us the transfer function of the NMOS LDO from the input to the output i.e. from the point V1 to Vo. V1 is the output of the subtractor circuit that performs the difference (Vo -βVref).The poles are the LHP Poles, The Resr Zero is a LHP zero whereas the fz 2 is a RHP Zero. 4.4 Pole Zero Locations for Full Load and No Load Conditions The targeted DC gain would decide on the values of transconductances gmx, gm1 and the output impedances R1,Rx. The load capacitor CL is known to us. Let us tabulate the poles and zeroes for the full load condition and the no load condition S.No Pole/Zero 1. fp1 First Pole 2. fz1 First Zero 4. fz2 Second Zero 5. No Load − gmp 2π CL −1 2π gmxRxR1Cc ∵ CL + gmpCxRx << gmpgmxRxCcR1 ∵ CL >> gmpCxRx + gmpgmxRxCcR1 gmpgmxCCR1 2π CLCx ∵ CL + gmpCxRx << gmpgmxRxCcR1 1 − 2π Re srCL 1 2π RxCx ∵ CL >> gmpCxRx + gmpgmxRxCcR1 1 − 2π Re srCL fp2 Second Pole 3. Full Load (500mA) fτ Unity Gain Bandwidth − − gmpgmxRx 2π Cc gm1 2π CC gmpgmxRx 2π Cc 1 2π gm1R1 gmxgmp CLCx Table (4.1) Pole-Zero Locations For Full Load and No Load 36 4.5 Choice of the Compensation Capacitor Cc Let us analyse a simple way of choosing the value of Cc that guarantees a good phase margin and then verify it with simulations. Based on the DC gain and the current consumption we target, we set the values of the transconductances and the output impedances. Let us consider the full load situation of 500mA and try to find out the value of the compensation capacitor. Let us target for a phase margin of 60°. The phase margin (Φ ) is given by the equation ⎛ fτ ⎞ ⎛ fτ ⎞ Φ = 180° − tan −1 ⎜ ⎟ − tan −1 ⎜ ⎟ → (5.6) ⎝ fp1 ⎠ ⎝ fp 2 ⎠ From [2] we know that for a phase margin of 60°, we need the second pole to be more than a factor of 2 more than the UGB. f p 2 >> 2 f τ 1 fτ ⇒ >> 2 fp 2 g m1 2π C L C x * 2π C C g mpg mxC C R 1 g m1C L C x ⇒ 0.5 >> g mpg mxC C 2 R 1 ⇒ 0.5 >> ⇒ C C >> 2 g m1C L C x → (5.7) g mpg mxR 1 4.6 Calculations of Poles and Zeroes Let us choose gm1=50µS, gmx=50µS, R1=48KΩ, Rx=120MΩ, gmp=2S Resr = 5mΩ, CL=470nF, Cx=50fF Our DC Gain and Cc are about Adc= gmxgm1R1Rx = 50e-6 * 50e-6 * 48e+3 * 120e+6 = 84dB Cc >> 2gm1CLCx 2*50e − 6*470e − 9*50e −15 = gmpgmxR1 2*50e − 6*48e + 3 >> 49*10e − 26 = 0.7 pF Therefore the minimum Cc we need is 0.7pF. So let us choose it as 1.7pF. 37 So our poles and zeroes turn out to be S.No Parameters 1. fp1 Full Load (500mA) 373Hz No Load 8.46Hz 52MHz 26KHz 9 MHz 9 MHz 3200G Hz 1.4GHz 5.7MHz 56KHz First Pole 2. fp2 Second Pole 3. fz1 First Zero 4. fz2 Second Zero 5. fτ Unity Gain Bandwidth Table (4.2) Pole-Zero Locations For The NMOS LDO ------ No Load ____ Full Load fp1 Gain 84dB fτ fp2 Frequency fz1 fp2 0° -45° -90° -135° -180° Frequency Figure (4.1a) Pole Locations Of The Compensated NMOS LDO 38 Hence we see that the NMOS LDO is stable after compensation. It has two poles and two zeroes. Under full load conditions the dominant pole comes from the error amplifier because of the Miller effect [1]. Under full load conditions, the NMOS LDO has a single pole response with its second pole located beyond the UGB. This is because the load pole is moved to gmpgmxCCR1 from the actual pole location of the NMOS Power transistor at 2π CLCx gmp . The main advantage of this compensation technique is that the active 2π CL feedback loop through the compensation capacitor buffers a lot of current even at KHz frequencies which gives enough gain to the LDO. This could also be seen as a shunt-shunt feedback active loop which offers a low output impedance. gmpgmxCCR1 from the actual pole In other words the output pole is moved to 2π CLCx gmp . This guarantees a phase location of the NMOS Power transistor at 2π CL margin of 90° at full load condition. Under no load condition we see that the dominant pole is contributed by the load capacitor. This means that the output of the LDO is same as the ground node after 8.46Hz and so the shunt-shunt feedback at the full load condition is not applicable to the no load condition. Also RxCx contributes another pole within the UGB. So at no load condition we have low phase margin (45°) unlike the full load condition. The Bode plot from the Spectre Simulation is shown in the Figure (4.2). 39 Figure (4.2) Bode Plot For Full Load and No Load Condition 4.7 Trade Off At No Load and Small Load Currents Since there are two poles within the UGB for small load currents, stability issue arises. Hence the LDO is brought back to stability by reducing the value of R x which moves the second pole beyond the UGB. This is evident from the table (4.1)because R x is the only term that is independent of the UGB at no load. This means that we are reducing our DC gain, A dc . We should only take care that the gain shouldn’t go lower than 30dB as it affects the accuracy of the output voltage. Then as soon as the load current starts increasing we don’t have an impact as the equations totally change with Miller effect dominating. 40 4.8 Implementation Of The Miller Compensated Circuit The circuit that is realized from the small signal model is shown in the Figure (4.3). It is the same schematic as the one shown in the Figure (3.6) with an additional Miller capacitor . The transistors MP135 and MP136 form the differential pair biased by a current source of 1.5µA. The reference voltage is supplied to the non-inverting terminal of the differential amplifier. The currents from the transistors MP135 and MP136 are mirrored through the NMOS current mirrors MN1, MN3 and MN2, MN4. Additional current sources have been added in MN1 and MN2 in order to control the output impedance of the error amplifier independent of the gm of the differential stage. The current in MN3 is mirrored across by the PMOS current mirror MP134, MP138. The drain of MP138 and MN4 is a high impedance point where we obtain the voltage of the error amplifier. The Miller compensation capacitor is connected from the drain of the MN3 transistor to the source of the NMOS power transistor. This is similar to the Nested Miller Compensation as mentioned in [1]. The output voltage of the error amplifier drives the level shifter. The output of the level shifter drives the NMOS power transistor. The source of the NMOS power transistor is connected to the resistive divider whose output voltage from a certain tap point is fed back to the inverting terminal of the differential amplifier. The source of the NMOS LDO which is our output node is connected to the pin via the bond pad and the bond wires. The Load capacitor is connected here. The load capacitor has a certain ESR and ESL of around 5mΩ and 2nH. The bond wires (two bondwires in parallel) have a resistance of 25mΩ to30mΩ and an inductance of 0.5nH. The resistance produces a zero with in the UGB which helps to improve the phase margin. This has a disadvantage that we have voltage drop of Iload*30 mΩ. The AC response and the transient characteristics are discussed in Section 4.9. 41 Figure (4.3) Realized Transistor Topology for NMOS LDO with Miller Compensation 42 4.9 AC Response And Transient Characteristics The open loop simulation has been performed to analyze the AC characteristics and the load step has been performed for load currents from 1mA to 500mA rising and falling in 1µs. The Figure (4.4) describes it all. Figure (4.4) AC characteristics of Miller Compensated NMOS LDO We see that the phase margin decreases from 90˚ to 30˚ from no load current till 1mA and then as the current starts increasing the phase margin increases back to 90˚. This is shown in the Figure (4.5). This is because for small load currents we have two poles with in the UGB and we could achieve 30˚ phase margin by reducing the DC gain of the LDO. Under small load conditions, the first pole comes from the load capacitor and the second pole comes from the error amplifier. The Miller capacitive multiplication doesn’t take place here because the load capacitor rolls of earlier 43 than the compensation capacitor and it is equivalent to connecting the compensation capacitor to ground. Plot of the phase margin and load currents on a logarithmic axis is shown below. Figure (4.5) Plot of Phase margin Plot vs Load Currents on a Logarithmic scale The transient response for a load step between 1mA to 500mA is shown in the Figure (4.6) below. There is some ringing observed when the load current increases to 500mA. This is observed for all temperatures and all process corners. The magnified version of the ringing is shown in the Figure (4.7). 44 Figure (4.6) Transient Response for a load step 1mA to 500mA Figure (4.7) Zoomed Version Of The Ringing 45 The frequency of the ringing was noted to be 2.5MHz which was initially thought to be the resonating feature of the LC tank at the load point. But the resonance frequency is 1 1 f = = = 3.6 MHz 2π ( Ltrack + LESL )CLOAD 2π 4e − 9 * 470e − 9 So the ringing is not due to resonance. A retrace back to the AC domain was made in order to analyse the cause for the ringing behaviour. A couple of issues associated with the ringing and the peaking seen in the Figure (4.7) are discussed in the Sections 4.10 & 4.11. 4.10 Ringing in Time Domain and Poor Gain Margin in AC Domain In AC domain for a particular load current of 50mA, the peaking in the gain is observed. This is because the non-dominant pole from the error amplifier clashes with the output pole of the power transistor for a particular range of load currents. This produces a complex pole (here we see that at 2.5MHz) that has a peaking in the gain with a 180˚ phase shift. This is shown in the Figure (4.8). In other words the damping at high frequencies is not sufficient enough to reduce this peaking. Figure (4.8) Gain Bump in Bode Plot @ Iout = 50mA. 46 4.11 High Frequency Peaking At high frequencies a strange peaking is observed. Along with the bondwire inductance and the ESL of the load capacitor, capacitance of the load capacitor, transcondutance of the power transistor and the parasitic capacitance at the source of the power transistor a tank circuit is formed . The parasitic capacitance for instance could be the Source to Nwell capacitance. This peaking is seen in hundreds of MHz range only for low load currents and is in several Giga Hertz range for the higher load currents that are outside our simulation range and are not shown in the Figure (4.4). The compensation of the NMOS LDO is still not yet complete because of the issue with poor gain margin for some load currents and the peaking at very high frequencies. Also we have a drop over the bondwire which could be eliminated. Improvements to overcome these problems are discussed in Chapter 5. 4.12 Summary In this chapter the Miller compensation technique of the NMOS LDO has been discussed. The advantage of this compensation is that we could move the load pole to higher frequencies from its actual location before compensation. We don’t need to rely on the ESR of the capacitor for the stability. However we have a trouble due to poor gain margin and are disturbed by the peaking in gain at very high frequencies. 4.13 References [1] Operational Amplifiers, Theory and Design by Johan.H.Huising, Kluwer Academic Publishers. [2] “A new method for multiplying the Miller capacitance using active components ,” by G.de Cremoux, Y.Christoforou, I.van Loo [3] “Optimized frequency-shaping circuit topologies for LDOs,” IEEE Trans. Circuit and Systems II, vol. 45, pp. 703–708, June 1998. [4] G. A. Rincon-Mora, “Active Capacitor Multiplier in Miller Compensated Circuits,”, IEEE J. Solid-State Circuits, vol.35, pp.26-32,Jan.2000. [5] Ka Nang Leung and Philip K.T.Mok , “A Capacitor-Free CMOS Low Dropout Regulator With Damping Factor Control Frequency Compensation,”, IEEE J. Solid-State Circuits, vol.38, pp.16911702,Oct.2003. [6] “ A Frequency Compensation Scheme for LDO Voltage Regulators,” by K.Chava and J.Silva-Martinez, IEEE J. Solid-State Circuits, vol.51, June 2004 [7] Embedded 5 V-to-3.3 V Voltage Regulator for Supplying Digital IC’s in 3.3 V CMOS Technology Gerrit W. den Besten and Bram Nauta, IEEE, J. Solid -State Circuits Vol. 33, No. 7, July 1998. 47 5 Damping Factor Enhancement, Power Sense Construction with Simulation Results ____________________________________________________________________ 5.1 Introduction The compensation technique discussed in Chapter 4 had the main disadvantage of obtaining a poor gain margin. The gain margin has to be improved, otherwise we will see oscillations during a closed loop operation of the NMOS LDO. The gain margin is poor because of the insufficient damping. A number of techniques have been proposed in order to improve the damping of the compensated operational amplifiers and LDOs [1-6]. The literatures mentioned above, suggest a common way to improve the damping factor by increasing the capacitance at the output of the error amplifier. Unfortunately in this architecture, this would have an impact on the stability at very low load currents say of the order of 100µA to 2mA. Hence a different approach has to be adopted in order to increase the damping ratio of the NMOS LDO. 5.2 Non-Dominant Poles The bump in the gain plot is observed because the non-dominant pole from the error amplifier clashes with the output non-dominant pole from the power transistor. As a result of this, for a particular current, the non-dominant poles clash thereby resulting in the peaking in the gain. The other way of seeing this is a poor damping factor. V1 Vref gm1V1 gmxVx gmL(VL-Vy) gmp(VL-Vo) X βVo Figure (5.1) Closed Loop In An Open Loop In open loop simulation, we open the loop at the “X” point. Despite opening the loop at “X”, we can still see that there is a closed loop through the 48 compensation capacitor Cc from the node Vo to Vx. When the gain in this internal closed loop is 1 with a phase shift of 180°, then the we can see the peaking in the gain or the Gain Bump is unavoidable. The damping factor can be calculated by opening the internal closed loop as discussed in [3]. In Figure (5.2) the internal closed loop is broken in order to analyze the damping factor. Vt gmxVx gmL(VL-Vy) gmp(VL-Vo) Figure (5.2) Open Loop Analysis Of The Internal Closed Loop Vo . This is now the open loop transfer function A( s ) β of our Inner Vt loop that we want to analyze. Solving now for 1 + A( s ) β = 0 we would be able to find out the damping coefficient and the frequency at which the peaking occurs. Solve for Vo − g mxg mL sC c ( g mp + sC p ) = A( s ) β = g mp g mL g oxg o1 + s ( C L g mL g ox ) + s 2 ( C L C xg o1 g mL + C L C cg oxg mL ) Vt Here g o1 = 1 1 and g ox = R1 Rx Simplifying further by neglecting the terms that are smaller than C Lg mLg ox , we get 1 + A( s ) β = 0 as g mpg mLg oxg o1 + s ( C Lg mLg ox ) + s 2 ( C LC xg o1 g mL + C LC cg oxg mL ) = 0 ⎡ ⎤ C Lg mLg ox g mpg mLg oxg o1 s+ ⇒ ⎢s2 + =0 C LC xg o1 g mL + C LC cg oxg mL C LC xg o1 g mL + C LC cg oxg mL ⎥⎦ ⎣ T his when compared with the standard equation s 2 + 2ξω ns + ω n 2 We get ω n = g mpg oxg o1 C L (C xg o1 + C cg ox ) and ξ= g oxg o1C L g mp (C xg o1 + C cg ox ) 49 The damping factor ‘ξ’ and the natural frequency ‘ωn’ have been derived and in order to increase the damping factor we have to decrease the open loop gain or increase the load capacitor or increase compensation capacitor ‘Cc’ or increase the transconductance of the power transistor. Unfortunately we can’t change the design parameters already fixed because of our targeted gain, smaller capacitor we want to use and the capability of the power transistor to handle 500mA load current. 5.3 Improved Damping Even though we don’t see the term Cp in the expressions of ξ, simulations confirm that ξ depends on Cp also. In the analysis done so far we have treated the LDO as a second order system but in reality we could relate the behavior of the LDO to a 4 th order system whose transfer function is very complex. Improvement in the damping factor could be achieved by connecting another capacitor (Cdamping) as shown in the Figure(5.3) between Vx and VL. V1 Vref gmxVx βVo gm1V1 gmL(VL-Vy) gmp(VL-Vo) Figure (5.3) Improved Frequency Compensation With the addition of the Cdamping capacitor we move the parasitic pole due to Cp to higher frequencies because of Miller effect. Now our first pole moves to lower frequency determined by gmxRxR1(Cc+Cdamping). In this design Cdamping capacitor of 800fF has been used to increase ‘ξ’ and remove the gain bump. Also in this design Cc+Cdamping ~ Cc and it is good that our UGB is not reduced further to hamper the transient response. The AC characteristic for 1.5V output along with the improved damping factor compensation is shown in the Figure (5.4). The transistor level schematic is shown in the Figure (5.5). 50 Figure (5.4) AC Characteristics with Improved Damping 5.4 Power Sense Construction We have seen that there is a drop of 15mV across the bondwire. We wish to avoid this drop by using an additional bondwire eventually demanding the need of an additional bondpad. The bondpads are named as Vldo_out and Vldo_regulate. This is called as the Power Sense Construction. The difference between this construction and the previous one is that the feedback in the Power Sense Construction is taken after the bondwires of the Power transistor. This is shown in the Figure (5.5). As a result of this we will not have a zero contributed by the bondwire to the stability of the LDO. But since our LDO is not dependant on this zero for stability, we don’t need to worry if we use the Power Sense construction. Topologies discussed in Chapter 3 suffer a lot in stability if we use the Power Sense Construction. 5.5 Peaking At High Frequencies At high frequencies we saw some peaking as explained in Section 4.11. This peaking is mainly seen for low load currents as the g mp is less and the peaking is because the damping in the path is poor. This is solved by adding a capacitor of 10pF in series with a 100Ω resistor from the source of the power transistor to ground. This is shown in the Figure (5.6). Also a resistor of 100Ω, has been added to the gates of the differential pair that helps to improve the damping in the tank formed by the bond wire in the feedback path along with the Cgs of the differential pair. This is important only when the feedback is taken directly from the Vldo_regulate node. (See section 6.1) This is shown in both the Figures (5.5) and (5.6). 51 Figure (5.5) Power Sense Construction Of The NMOS LDO 52 Figure (5.6) Additional RC circuit to remove the peaking at High Frequencies 53 5.6 AC Characteristics The AC response for the final schematic shown in Figure (5.6) is shown below. The phase margin is plotted as a separate part. They are shown in the Figure (5.6) and (5.7) respectively. We can see that the peaking at high frequencies disappeared. Figure (5.6) AC Response with Power Sense Construction & Improved Compensation Figure (5.7) Phase Margin Plot for the AC Response 54 5.7 Transient Response The transient response of the NMOS LDO with the power sense construction is shown in the Figure (5.8). The voltage drop of 15mV is not seen anymore. The load step is between 1mA to 500mA. Figure (5.8) Transient Response Of the NMOS LDO 5.8 DC Response The DC simulation is performed by a DC sweep on the regulator input Vldo_in. We measure the regulator output Vldo_out. The Drop Out voltage is a difference between Vldo_in and Vldo_out when the output voltage drops to 1% of its value. This is shown in the Figure(5.9) where Vldo_in, Vldo_out and their difference is plotted. Also a DC sweep of load current from no load till 1A was swept in order to observe the maximum current handling capability of the NMOS LDO. It is seen in the Figure (5.10), that the output voltage Vldo_out drops to 1.475V when the load current is 1A. This means that the % error is only 1.6%. The DC load regulation is calculated for a current load from 1mA to 500mA from the Figure (5.9) to be % DC Load regulation = 1.495 − 1.492 *100 = 0.6% (500 − 1)e − 3 which is within our specification of 1% (See Table 1.1, Section 1.7) 55 Drop Out Voltage=160mV Figure(5.9) DC Sweep and the Drop-Out Voltage Simulation Figure (5.10) LDO Output voltage vs Load Current 56 5.9 PSRR Calculations The Power Supply Rejection Ratio is defined as the ratio of the variation in the output voltage to the variation in the supply voltage. The variation is modelled as a small signal in the supply Vldo_in and VBAT . The variation in the output voltage is measured and the output voltage (Vldo_out) in decibels is plotted. Ideally for a NMOS LDO, the PSRR is infinity because the NMOS power transistor is biased in saturation region and any variation in Vldo_in doesn’t change its current, provided the error amplifier has no variations. For 1.5V output scenario at 500mA load current and 470nF load capacitor, the PSRR at 1MHz is –32dB. This is shown in the Figure (5.11). Figure(5.11) PSRR of the NMOS LDO 5.10 Monte-Carlo Simulations The transistors have been dimensioned in such a way that the mismatch spread of these transistors should result in a 4σ spread of 2%. This is possible only if the W.L is big enough according to [6]. At the same time bigger W.L results in a bigger parasitic capacitances and could hamper the speed of the LDO. Hence optimisation is done by analysing the spread by running Monte-Carlo simulations and then re-running the AC and Transient simulations. The Histogram indicating the spread in terms of σ is shown in the Figure (5.12).The simulation has been performed for lower voltage ranges of VISA and VBAT. From this σ value, the mismatch spread in terms of percentage is calculated. 57 4σ *100 X where X is the mean value and σ is the standard deviation. % spread = Here we get σ as 11.6mV and X = 1.4759V 4*11.6e − 3 % spread = *100 = 3% 1.4759 Figure(5.12) % Spread of the Output Voltage of the NMOS LDO 5.11 Equivalent Output Noise The equivalent output noise is also small in µV per sqrt(Hz) range at low frequencies with the noise decreasing at high frequencies. This is shown in Figure(5.13). The equivalent output noise is a measure of the performance of the LDO. This can also be considered as a Figure Of Merit of the LDO(FoM). Lower the noise output, higher the FoM. 58 Figure (5.13) Equivalent Output Noise of the NMOS LDO 5.12 Summary The improved frequency compensation of the NMOS LDO has been discussed in this chapter. The issues associated with the LDO in Chapter 4 have been solved and verified through simulations. The Power Sense Construction that enables the LDO output to have an output voltage error of less than 1%has been discussed. Very important to be noted is that the load capacitor should be placed as close as possible to the pin. Placing away results in larger inductance which would result in a higher gain at high frequencies when the phase would be greater than 180° resulting in oscillations at the output. 5.13 References [1] Analog Circuit Design by J.Huising, R.J van der Plassche and W.Sansen [2] Advanced Low Voltage and High Speed Techniques for BiCMOS, CMOS and Bipolar Operational Amplifiers, by K.J.de Langen, Kluwer Publishers. Ph.D Thesis, TU Delft. [3] Microelectronic Circuits, 5th edition, Sedra/Smith. [4] Design Of Analog Integrated Circuits & Systems, Kenneth.R.Laker and Willem.M.C.Sansen. [5] Positive feedback Compensation in Operational Amplifiers by J.Ramos and M.Stayeart. [6] Analogue-Digital ASICs circuit techniques,design tools and applications, Edited by R.S.Soin,F.Maloberti and J.Franca, Peter Peregrinus Ltd for IEEE. 59 6Additional Features In The NMOSLDO _____________________________________________________________________ 6.1 Programmability of the output voltage The NMOS LDO has been designed for 1.5V output. The output of the NMOS LDO can be changed by changing the feedback point. This is explained in the Figure (6.1) with a simple NMOS LDO structure. Vref VLdo_in + - VLdo_out Ra IL CL Rb Figure (6.1) Programmable Output NMOS LDO The value of Ra can be varied and so the feedback factor ‘β’ which in turn generates a different output voltage. In the design discussed here so far the output could be programmed from 1.5V down to 0.6V. Every tap point is connected through a switch to the inverting terminal of the error amplifier. The switches will be selected by means of decoder. The number of inputs to the decoder depends upon the number of programmable taps we would like to have. We have the worst case feedback when ‘β’=1. We open the loop during the open loop AC simulation (Section 3.6). Even though the parameters of the transistors remain the same, the open loop gain and the UGB of the LDO increase when the output voltage is programmed from 1.5V to 0.6V. Hence the worst case situation of the LDO to guarantee stability is the 0.6V scenario. The AC characteristics for the 0.6V output are shown in the Figure (6.2) and the phase margin is plotted versus load currents separately in the Figure (6.3). The phase margin is 90° for larger load currents and no load situation. A decrease in phase margin close to 25° is observed for smaller 60 currents around 200µA. The phase margin then increases towards 90° for larger load currents. Figure (6.2) AC Characteristics for 0.6V output Figure (6.3) Phase Margin at 0.6V output 61 The transient characteristics of the 0.6V output for a load step from 1mA to 500mA is shown in the Figure (6.4). Figure (6.4) Transient Output for 0.6V 6.2 ECO Mode There always arises a situation when the load current can reduce to 15mA. This mode is called as the ECO mode or the Stand By Mode of the applications like the Cellular Phones, Apple IPODs etc. In this mode the expected quiescent current is only of the order of the 15µA and the stability should be guaranteed. The design approach is straight forward by reducing the bias currents to the error amplifier and the level shifter. This is shown in the Figure (6.5) for 0.6V output. Also the simulation results are shown for the 0.6V output. The AC response for the ECO mode is shown in the Figure (6.6). The phase margin plot is shown in the Figure (6.7). If the load current in the ECO mode increases to 40mA then we need to increase the C damping capacitor or the bias currents in order to avoid the gain bump. The first solution is preferred in order to stay within 15µA quiescent current consumption. 62 Figure (6.5) ECO Mode Implementation 63 Figure (6.6) ECO Mode: AC Characteristics at 0.6V Figure (6.7) Phase Margin in ECO Mode for 0.6V output 64 The transient characteristics for a load step of 100µA to 20mA at 0.6V output is shown in the Figure (6.8). Figure (6.8) Transient Response in ECO Mode 6.3 Adaptability Of The LDO To Scalable Load Currents The NMOS LDO shown in the Figure (5.5) delivers a programmable voltage. The resistive divider is made programmable by changing the feedback point across the Resistors .i.e. ‘β’.The second feature is the adaptability of the NMOS LDO. The LDO in the Figure (4.5) handles a maximum current of 500mA. In case the maximum current is set to 250mA or even lower the LDO can be adapted just by scaling down the dimension of the power transistor and the load capacitor accordingly. When we decrease the current by a factor of 2, we also decrease the dimension of the power transistor by a factor of 2 and the load capacitor by a factor of 2. This means that in the AC domain our poles remain the same as the scaling is proportionate. This is shown in Table 6.1. Since the pole locations remain the same and the UGB unchanged, the overshoot and undershoot remain the same in the transient domain. The zeroes fall at very high frequencies and have no impact on the performance of the LDO. This means that if we use 470nF for a load of 500mA, we can scale down our load capacitor accordingly for lower load currents as shown in Table 6.2. 65 Table 6.1 Impact of Poles and UGB with Load Current Scaling S.No Load Currents 1. ILoad First Pole UGB Second Pole 1 2π gmxRxR1Cc gm1 2π CC gmpgmxCCR1 2π CLCx 2. 500mA 1 2π gmxRxR1Cc gm1 2π CC gmpgmxCCR1 2π CLCx 3. 350mA 250mA Independent of the load Current Independent of the load Current 0.7 * gmpgmxCCR1 = Same as 500 mA condition 2π CL * 0.7Cx 4. Independent of the load Current Independent of the load Current 5. 150mA Independent of the load Current Independent of the load Current 6. 100mA Independent of the load Current Independent of the load Current 7. 50mA Independent of the load Current Independent of the load Current g mp gmxC CR1 2 = Same as 500 mA condition CL 2π Cx 2 g mp g mxC C R 1 3 = Same as 500 mA condition CL 2π Cx 3 g mp g mxC C R 1 5 = Same as 500 mA condition CL 2π Cx 5 gmp gmxCCR1 10 = Same as 500mA condition CL 2π Cx 10 Table 6.2 Load Currents Vs Load Capacitors S.No Load Current Load Capacitor 1. 500mA 470nF 2. 350mA 330nF 3. 250mA 235nF (220nF) 4. 150mA 150nF 5. 100mA 94nF (100nF) 6. 50mA 47nF 66 6.4 Cap Free Option gmp , we 2π CL could also connect a wide range of capacitors for the LDO. This is called as the cap free option. From calculations, we see that we can connect any load capacitor between 470nF to 470µF. This still will make the output pole to be beyond the UGB. Connecting the maximum capacitor will have a better phase margin at no load, medium load current situation. The dominant pole is from the Load capacitor for a range of currents from 0 to 10mA and as the current increases, the Miller capacitor dominates and the phase Margin at full load is around 45°. This is because the second pole has moved closer to the UGB such that the phase margin is 45°. The phase margin plot (vs) the load current on a logarithmic scale is shown in the Figure (6.9). Since we see that the output pole moves beyond its limiting value Figure (6.9) Phase margin Vs Load Current with a 470µF load capacitor 6.5 Summary The additional features like the programmability of the output voltage, ability to use the same NMOS LDO in the ECO mode and the adaptability of the NMOS LDO transistor dimensions and load capacitors when load currents scale down were discussed. 67 7 Conclusions And Future Work _____________________________________________________________________ The thesis is about the design techniques of an NMOS power transistor based LDO. In Chapter 1, the PMU was introduced along with the need for the NMOS LDOs. In Chapter 2, the design approaches for the PMOS LDOs were discussed in order to verify if the same approach can be used while designing the NMOS LDOs. In Chapter 3, the conventional design of the NMOS LDOs were discussed, along with the problems associated while adapting the LDO to scalable load currents. In Chapters 4 and 5, the motivation for the frequency compensation of the NMOS LDO along with the improved damping technique was discussed. In Chapter 6, the additional features related to the NMOS LDO like its programmable output voltage, ECO mode and load adaptability were discussed. In Chapter 7, a summary of the first 6 chapters has been presented in nutshell. Future Work Even though this NMOS LDO design has met the specifications that are needed for an LDO to be used in a mobile PMU, there is always a scope for improvement. The major improvement should be to improve the phase margin of the NMOS LDO at low load currents. In this design the phase margin is about 25° to 30° for load currents between 200µA to 2mA. The phase margin should be improved for these currents. Also the NMOS LDO could be made faster or the UGB can be increased in order to achieve the best transient response. 68 Appendix- I Small Signal Analysis Of The Compensated NMOSLDO Writing the nodal equations At the node Vx gm1V 1 + Vx + (Vx − Vo)sCc = 0 → (1) R1 ⇒ gm1V 1 = Vx(1 + sCcR1) ⇒ Vx = sCcVo − gm1V 1 → (2) 1 + sCcR1 Assuming C1<Cc. We neglect C1 to simplify the calculations At the node Vy gmxVx + Vy (1 + sCxRx) = 0 → (3) Rx ⇒ Vy = − gmxVxRx → (4) 1 + sCxRx At the level shifter node VL VL gmL (VL − Vy ) + + (VL − Vo) sCp = 0 → (5) RLS 1 ⎞ ⎛ ⇒ ⎜ gmL + sCp + ⎟ VL − gmLVy = sCpVo → (6) RLS ⎠ ⎝ Since RL ~ 1/gmLS equation (6) gets reduced to (2 g m L + sC p )V L − g m L Vy = sC pVo (2g m L or g m L doesn't m ake m uch difference here) ⇒ ( g m L + sC p )V L − g m L Vy = sC pV o ⇒ VL = ⇒ VL = g m L Vy + sC pVo g m L + sC p − gmL g m xV xR x + sC pVo 1 + sC x R x → (7) g m L + sC p 69 At the output node Vo we get (V x − V o ) sC c + (V L − V o ) sC p = g mp (V o − V L ) + ⇒ (V x − V o ) sC c + (V L − V o ) sC p = Vo 1 ⎞ ⎛ ⎜ R e sr + ⎟ sC L ⎠ ⎝ sC L V o + g mp V o − g mp V L sC L R e sr + 1 sC L ⎛ ⎞ ⇒ sC cV x + ( g mp + sC p )V L = V o ⎜ + g mp ⎟ ⎝ sC L R e sr + 1 ⎠ ∵ C c , C p << C L g mxV xR x ⎡ ⎛ ⎞⎤ ⎢ ⎜ − g mL 1 + sC xR x + sC p V o ⎟ ⎥ sC L ⎛ ⎞ ⇒ sC cV x + ⎢ ( g mp + sC p ) ⎜ + g mp ⎟ ⎟⎥ = Vo ⎜ g mL + sC p ⎝ sC L R e sr + 1 ⎠ ⎢ ⎜ ⎟⎥ ⎝ ⎠ ⎦⎥ ⎣⎢ sC p << g mp & sC p << g mL ⎡ ⎛ g mp ⎞ ⎛ g mxV xR x sC L ⎞⎤ ⎛ ⎞ ⇒ sC cV x + ⎢ ⎜ + sC p V o ⎟ ⎥ = V o ⎜ + g mp ⎟ ⎟ ⎜ − g mL 1 + sC xR x ⎠⎦ ⎝ sC L R e sr + 1 ⎠ ⎣ ⎝ g mL ⎠ ⎝ ⎛ g mp g mxR x ⎤ sC L g mp sC p ⎞ ⎡ ⇒ V x ⎢ sC c − = Vo ⎜ + g mp − ⎟ ⎥ e 1 R 1 + sC R sC + g mL ⎠ x x L sr ⎣ ⎦ ⎝ g mp g mxR x ⎞ ⎛ ( sC cV o − g m 1V 1) R 1 ⎞ ⎛ ⇒⎜ ⎟ ⎜ sC c − ⎟ 1 + sC cR 1 1 + sC xR x ⎠ ⎝ ⎠⎝ ⎛ sC L g mL + g mp g mL ( sC L R e sr + 1) − g mp sCp ( sC L R e sr + 1) ⎞ = Vo ⎜ ⎟ g mL ( sC L R e sr + 1) ⎝ ⎠ ⎡ s 2 ( C c ) 2 R 1 − sC cg mxg mp R xR 1 { g mp g mL ( sC L R e sr + 1 ) − sC p g mp ( sC L R e sr + 1 ) + s C L g mL} ⎤ ⎥ ⇒ Vo ⎢ − g mL ( sC L R e sr + 1 ) ⎢⎣ (1 + sC xR x )(1 + sC cR 1 ) ⎥⎦ g m 1V 1 R 1 ⎡⎣ sC c (1 + sC xR x ) − g mxg mp R x ⎤⎦ = → (8) (1 + sC cR 1)(1 + sC xR x ) In the above calculations, VL has been expressed in terms of Vx. Refer to equation (7). The s2 terms can be neglected in order to make the calculations simpler and By substituting similar values for the parameters like the previous Calculations we infer that, the s 2 terms will have an impact only at frequencies in the Giga Hertz range. Still simplifying the equation (8) we get approximately 70 ⎡ − gmxgmpRxsCcR1 {gmpgmL(1 + sCL Re sr ) + sCLgmL)} ⎤ gm1V 1R1 [ sCc (1 + sCxRx) − gmxgmpRx ] Vo ⎢ − ⎥= (1 + sCxRx )(1 + sCcR1) gmL(1 + sCL Re sr ) ⎣ (1 + sCxRx )(1 + sCcR1) ⎦ ⇒ = Vo [ − gmxgmpRxsCcR1 gmL (1 + sCL Re sr ) − {{gmpgmL (1 + sCL Re sr ) + sCLgmL)}(1 + sCxRx )(1 + sCcR1)}] (1 + sCxRx )(1 + sCcR1) gmL (1 + sCL Re sr ) gm1V 1R1 [ sCc (1 + sCxRx ) − gmxgmpRx ] (1 + sCxRx )(1 + sCcR1) ⇒ gm1 gmLR1 [ sCc(1 + sCxRx ) − gmxgmpRx ][1 + sCL Re sr ] Vo = V 1 [− gmxgmpRxsCcR1 gmL(1 + sCL Re sr ) − {{gmpgmL (1 + sCL Re sr ) + sCLgmL)}(1 + sCxRx )(1 + sCcR1)}] ⎡ sCc ⎤ gm1R1 gmxRx ⎢ − gmp ⎥ [1 + sCL Re sr ] Vo ⎣ gmxRx ⎦ ⇒ = → (9) V 1 [− gmxgmpRxsCcR1(1 + sCL Re sr ) − {{gmp (1 + sCL Re sr ) + sCL}(1 + sCxRx )(1 + sCcR1)}] (∵ sCc >> s 2CcCxRx ) Equation (9) highlights us the transfer function of the NMOS LDO which has been derived from the basic model as in fig(1). Poles and Zeroes The numerator in the transfer function when equated to zero results in the “Zeroes” of the transfer function while the denominator when equated to zero results in the “poles” of the transfer function. Starting with the poles we equate the denominator of the equation to the standard format s s2 + → (10) p1 p1 p 2 with the assumption that p1 is dominant that p 2. p1 and p 2 are the two poles. 1+ By neglecting the s 2 terms and the sC L R esr term we can simplify the denominator of equation(9) as ⎛ CL ⎞ s 2CLCxRx 1+ s ⎜ + CxRx + gmxRxCcR1 ⎟ + → (11) gmp ⎝ gmp ⎠ 71 Equating the ‘s’ and the ‘s2’ terms we can get the expressions for the first and the second pole. 1 CL = + CxRx + gmxRxCcR1 p1 gmp ⇒ p1 = gmp CL + (CxRx + gmxRxCcR1) gmp Similarly ⇒ p2 = p1 p 2 = → (12) gmp CLCxRx ( CL + gmpCxRx + gmxgmpRxCcR1) → (13) gmp = CLCxRxp1 CLCxRx Now we look at the zeroes in the transfer function by equating the numerator to Zero with an assumption that sCc << s2CcCxRx ⇒ gmLgm1( s 2C cC xRx − gmpgmxRx ) = 0 ⇒ fz1 ~ gmpgmx 2π C cC x (1 + sC L R e sr ) = 0 −1 ⇒ fz 2 = − 2π R e srC L → (14) → (15) The zero due to the compensation capacitor is located at high frequencies in the order of hundreds of Giga Hertz range at no load to thousands of Giga Hertz in the full load condition. Also we have a Left Hand Plane Zero due to the load capacitor and its equivalent series resistance. A single pole system is unconditionally stable. At low load situations we have a two pole system. But the second pole can be moved by reducing the output impedance of the amplifier i.e. the value Rx. 72 73 Appendix- II Load Step Simulations from 0 to 500mA We saw in the previous sections, that the transient step was from 1mA to 500mA which means that the minimum current load is 1mA and the maximum load current is 500mA. Nevertheless the minimum load current can be zero, say for example the load gets switched off or in real life situation when the LDO is powered first before the load. Very important is that in real life situations the LDO never sees a step of 500mA as soon as the load is turned ON. Over a span of time the maximum load current is 500mA. This is shown in the Figure below. LDO Sum of the Load Currents is 500mA CLOAD The load current in most of the applications reaches its maximum value of 500mA in a staircase fashion. 500mA Load Current 0 mA Time in µs Load Current Reaching its Maximum Value 74 The transient simulation for such a staircase load step is shown below for the 1.5V and 0.6V voltage outputs. Transient Response For A Staircase Load Current Step 75 Also the transient simulation for the load step from 0 to 500mA is shown below for 1.5V and 0.6V output voltages. Here the overshoot and undershoot are high, even around 7.5% because the LDO is not fast for load currents less than 100µA. The duration of the output voltage peaks and dips are 4ms. LDO Output Voltage For A 0 to 500mA, Load Step rising and falling in 1µs 76