A Fully-Integrated Self-Healing Power Amplifier

A Fully-Integrated Self-Healing Power Amplifier
of 4
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  A Fully-Integrated Self-Healing Power Amplifier  Steven M. Bowers, Kaushik Sengupta, Kaushik Dasgupta, and Ali HajimiriCalifornia Institute of Technology, Pasadena, California, 91125, USA Abstract —The first fully-integrated self-healing mm-wavepower amplifier heals process variation, load mismatch andtransistor failure with on-chip sensors, actuators and anintegrated digital algorithm ASIC without externalcalibration. Measurements of 20 chips showed increased RFpower up to 3dB, or reduced DC power by 50% in backoff at28 GHz. Healing 4-1 VSWR load mismatch for RF and DCpower improvement was verified, and healing after laserinduced transistor failure increased RF power up to 4.8dB. Index Terms — Actuators, CMOS Integrated Circuits,Digital Control, Power Amplifiers, Power Generation,Sensors, Thermal Sensors.   I. I  NTRODUCTION   Silicon mm-wave integrated circuits have enablednumerous new high frequency applications that previouslywere not economically feasible. However, high power mm-wave design in nanometer scale CMOS technologiesis particularly susceptible to transistor variations, modelaccuracy, environmental changes, and aging [1]. Reliablemodeling is also a greater challenge in high frequencycircuit design as the transistor parasitics contribute a larger fraction of the total relevant circuit capacitance. This isfurther exacerbated by the fact the models are often primarily optimized for digital circuits. Finally, agingeffects and environmental variations such as loadmismatch and temperature can significantly degrade the performance of mm-wave power amplifiers (PAs), where both output power and efficiency are strongly dependenton process and environmental parameters [2]. Self healingcan mitigate these issues by identifying any degradationand modifying the parameters of the circuit to improve its performance post fabrication with minimum overhead[3,4]. In this work, integrated self healing of a PA isaccomplished by using on-chip sensors, actuators, analogto digital converters (ADCs), digital to analog converters(DACs), and a digital application-specific integratedcircuit (ASIC) providing a self-contained self-healingsystem.This paper demonstrates a mm-wave PA system (Fig. 1)where on-chip self-healing improves the output power,gain, and DC power consumption under a broad This work was supported by the Air Force Research Laboratory.The views expressed are those of the authors and do not reflect theofficial policy of the Department of Defense or the U.S. Government. Fig. 1. Block diagram of self-healing power amplifier showing RF path, sensors, actuators, data converters and digitalcore. range of non-idealities, such as process variations,modeling inaccuracies, load mismatch, and evencatastrophic failure of amplifier. It is essential that theadditional self-healing circuitry introduce minimaldegradation of the RF performance with a low overhead interms of area and power consumption. The individualactuator, sensor and data conversion blocks are presentedin [5]. A sufficient actuation range is required toencompass the optimum state for all expected variations.Similarly, sensors robust to the very same variations theyare trying to detect are also necessary and are achieved byusing non-minimum length transistors and variation-insensitive topologies [5]. While previous work has shownaspects of off-chip healing, [3,4] to our knowledge this isthe first demonstration of a fully integrated self-healingmm-wave system.II. D ESIGN   The self-healing power amplifier (Fig. 1) combines the power from two amplification stages using a 2:1transmission-line-based output-combining network. Eachamplifying path consists of two cascode stages, wherethick gate-oxide common-gate transistors are used toimprove gain and high voltage handling capability. Digitized Sensor DataActuation States CMOS CHIP RF IN ADCADC RF OUT RF PowerSensor ADC DC CurrentSensor RUN HEALING T-line ActuatorT-line ActuatorRF PowerSensorDC CurrentSensorTempSensorTempSensorTempSensorBias DACDC CurrentSensorTempSensorDC CurrentSensorSelf HealingCoreBias DACBias DACBias DAC  A. Actuators  The output matching network can be tuned though threetunable transmission line stub actuators (Fig. 2) that are programmed with digital switches that short the signal lineto the local ground at various points [5]. This is especiallyuseful for healing against load mismatches, as shown inthe measurement plot of Figure 2, where the total output power of the PA was measured under two different loadimpedances while the actuator state of one of thetransmission line stubs was swept. Additionally, gatevoltages of the eight amplifier transistors are controlledindependently by 6-bit low-power bias actuation DACs,enabling 5x actuation range of the DC power. Fig. 2. Schematic of transmission line actuators and performance variation for different actuator states under differentload mismatch. B. Sensors and ADC  Low power, yet accurate sensors are essential to theself-healing system. The magnitudes of the input andoutput RF power are measured using two 20dB directionalcouplers with power detectors both at the coupled andisolated ports (Fig. 1). This facilitates detection of wavestraveling both to the load and reflected from it, enablingaccurate power sensing in the presence of load mismatch.The four supply voltage regulators associated with eachamplification stage are modified to also serve as DCcurrent sensors (Fig. 3). Junction temperature sensorsutilizing interdigitated diodes between fingers of theamplifying transistors assess temperature differences between the transistor cores and other points on the chip,as a measure of the DC power consumption. On-chip 8-bitADC’s relay the sensor data to an integrated digital ASIC.Due to their robust design, no external calibration isneeded or used in any actuator, sensor, DAC, or ADC [5]. Fig. 3. Schematic of a single stage of the amplifier showingthe DAC biasing, thermal sensor diodes, and voltage regulator with DC current sensor. C. Digital Algorithm The synthesized on-chip healing algorithm is made upof a global state machine that controls component blocksfor actuation load, sensor read, and optimization. Becauseof this modular code setup, many different types of complex optimization algorithms can be incorporated intothis general fully-integrated self-healing framework. Twosuch optimization strategies have been synthesized in thecurrent implementation that finds the optimum solutionwithin the search space of 262,144 possible states. Onefinds the state with maximum output power, and the other finds the state with the lowest DC power for a givenoutput power. The convergence of the algorithm using alow frequency clock of 50MHz (limited by themeasurement setup) takes up to a maximum of 0.8seconds. At its full speed of 200 MHz, the maximumhealing time would be 0.2 seconds.III. M EASUREMENT   The self-healing PA system is fabricated in a 45nm SOICMOS process and 20 chips were evaluated under nominal loads, 10 of which were also measured withmismatched load conditions generated by a variable mm-wave load tuner (Focus Microwaves 50GHz iTuner).Using simulations, actuator settings that maximize performance metrics of interest (e.g., output power) atsaturation have been determined for typical transistors andare defined in measurements as “un-healed” default states.Because the default state was designed for saturation, it iscloser to optimum than at lower input powers, and thus theeffectiveness of the healing is greater in back off conditions. After healing, the PA has a small signal gainof 21.5dB and a power added efficiency of 7.2% whiledrawing 520mW of DC power at 28GHz. Fig 4. showsoutput vs. input power in the default state as well as in the 123456781111.51212.513T-line actuator setting    O  u   t  p  u   t   P  o  w  e  r   (   d   B  m   )   Z L = 50*(0.565 + 1.04j)  Z L = 50*(0.542 - 1.03j)    B  i     a s B  i     a s   states healed at small signal and near compression. Itshows that there is no one state that can provide theoptimal performance at all input powers, but with self healing, the appropriate state for the current environmentis automatically found. The two histograms in Fig. 4   show the improvement of 20 chips when healed in back-off and near compression. Fig. 4. Output power healing from default state to healedstates at small signal and near compression with 20 chiphistograms showing output power improvement in back-off andcompression. Using the on-chip algorithm to minimize DC power while maintaining a given output power, the DC power  required to reach different output RF power levels with aminimum of 10dB gain is plotted for 20 chips and thecross section at Pout= 12.5dBm is shown as a histogram inFig. 5. Fig. 5. DC power healing shown vs output power, with a 20chip histogram showing variation at P out =12.5dBm Another critical benefit is that variation between chips isalso significantly decreased with self healing, as seen inFig. 5.To demonstrate self healing under load mismatch, theon-chip algorithm is run for various VSWR points up tothe 4:1 circle. Fig. 6a shows constant output power contours on a 50 Ω Smith chart for 0.7dBm input power,under different load impedances using default settings aswell as with self healing, showing dramatic improvementacross all of the measured loads. The improvement across10 chips for 2 specific load points is plotted as histogramsin Fig. 6b.   Fig. 6. a) Output power healing under load mismatch showingself-healing operation within the entire 4-1 VSWR circle. b) 10chip histograms showing healing ability at 2 loads. c) DC power reduction while maintaining 12.5dBm output power. One key advantage of this self-healing system is that itcan provide a constant output power across VSWR eventswhile simultaneously reducing the DC power consumption. To showcase this ability, the on-chipalgorithm is instructed to keep 12.5dBm output power with minimum power consumption across all loads. Asdepicted in Fig. 6c, the healed PA is able to cut DC power levels by up to 33%. -15-10-50551015Input Power (dBm)    O  u   t  p  u   t   P  o  w  e  r   (   d   B  m   )   without self-healingself-healing at small signalself-healing at 1dB CP 789101112132468Output Power (dBm)   without self-healingwith self-healing   13.51414.51515.51616.517246810Output Power (dBm)   without self-healingwith self-healing   891011121314151617100200300400500600700 P OUT (dBm)    D   C  p  o  w  e  r  c  o  n  s  u  m  p   t   i  o  n   (  m   W   ) without self-healingwith self-healing 20025030035040045050055060065051015DC Power (mW)   with self-healingwithout self-healing   101112131415 1   3   . 2   5    1   2   . 5    12131415161712345Output Power (dBm)   without self-healingwith self-healing   10111213141234Output Power (dBm)   without self-healingwith self-healing     350400450500550600   3    5     0    4     0    0      The linear involving nomagnitude (also measurehistogram of  presented infrom the fac power cancompression.system to op Fig. 7 EVand 100ksps 1  Finally, selfailure by laamplifier stataken after hwere cut outransistor wawas cut out,the on-chip4.8dB, and 3healing partiAs the defaulof the system Fig. 8 Outptransistor failu 212345   -15-10-505101520Input    O  u   t  p  u   t   P  o  w  e  r   (   d   B  m   )   -15-10-505101520Input    O  u   t  p  u   t   P  o  w  e  r   (   d   B  m   )   ity of the P-constant enVM) in bothd under a 10010 chips for Fig. 7. The ithat with bee achievedThis alsorate as either  histogram for QAM modulati f healing wasser cutting pes. As showalf of one of , and then as cut. Finallealing an 8 dealing algorit.4dB at saturaularly under t state goes futo heal increa ut power vs. ie for default an   345E -505Power (dBm)without self-healinwith self-healing-505Power (dBm)without self-healinwith self-healing is criticalelope. Thus,default and hsps 16QAMn output pow provement iter matching,t a lower lehighlights thclass A or cl 10 chips at 12.5on before and a applied in therts of one of in Fig. 8, mehe common sain after hal, that entire blow to thehm was ableion, showingextreme brearther from optises, even in sa nput power atd after self heali 67M (%)withoutwith sel -15-10-505101520In    O  u   t  p  u   t   P  o  w  e  r   (   d   B  m   )   g   -15-10-505101520In    O  u   t  p  u   t   P  o  w  e  r   (   d   B  m   )   g for applicatiothe error vecaled states wodulation anr of 12.5dBmlinearity cothe same outel of transisability of tss AB amplifi  dBm output poter healing.  case of transisthe two outasurements wource transistof the cascamplifying stefault state, to heal by uphe utility of sdown scenarimum, the abiluration. various pointsng. 89   self-healingf-healing -505put Power (dBm)without self-hewith self-healin-505put Power (dBm)without self-hewith self-healin nstor reaisesuttor iser. er  tor utrersdegeattoelf os.ity of Fig. 9 blocks,up phot This benefithealingamplifisignifiDC po procesmeasur mismaThedigital [1] K.S. per  be50,[2] A.HB  Th no.[3] J.H.-selco Ci 5-7[4] L.trano.[5] K.“OSel Su     alingg   alingg Die photo osensors, actuatoos of output sta  paper dems and the flexsystem wither. Two onant increaseswer for a givvariation aements fromch and transisAauthors wouldsynthesis supp Bernstein, D.J.. Nassif, E.J.formance CMond,” IBM Jo no. 4.5, pp. 43Keerti, A.-V.T power ampl eory and Tech .2, pp.207-214,.-C. Liu, A. TH. Hsieh, W. Pf-healing powetrol in 65 nm cuits Symposiu June 2011Sankey, Z. Pnsmitters," Mic ., pp.225-228, 7Sengupta, K.n-Chip Sensinf-Healing m bmitted to Int   f self-healingrs and self-heale transistor bef  II.   C ONCLUSI nstrates man bility providea mm-wave c-chip healingin output RFn output RFnd mismatch0 chips, andtor failure was KNOWLEDGE like to thank ort.R  EFERENCES Frank, A.E. Gatowak, D.J. PeaS variability i rnal of Resear -449, July 2006. Pham, “RFfiers under loa niques, IEEE T eb. 2007ng, N.-Y. Wano-Yi, C. Jou, Mr amplifier witCMOS," Radm(RFIC), 201 opovic, "Adapt rowave Sympo 12 June 2009Dasgupta, S. Band Actuatio-Wave CM rnational Micr A system shng digital core,re and after las  NS  y of the sd by integratiircuit such asalgorithms a power or decower. Healinwere verifiealing load idemonstrated ENT  Ben Parker of  tiker, W Haensc   rson, N.J. Rohr the 65-nm re h and Develop .characterizatiod mismatch,” ransactions on g, Q.J Gu, R..-C.F Chang, "adaptive feed io Frequency IIEEE , vol., nve tuning for  siumDigest, 2 owers, and A.Methods for S Power  owave Symposi wing RFand closeer cutting.   bstantialg a self-a power llow for reases ing againsted with pedance.IBM for  h, B.L. Ji,er, “High-gime and ment,   vol.of SiGe icrowave , vol.55,erenguer,A V-band back bias Integrated o., pp.1-4,handheld 009 , vol.,Hajimiri,Integrated   mplifier” ium2012
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks