mixed-signal socs with in situ self-healing circuitry

Upload: prasanth-varasala

Post on 03-Apr-2018

215 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/28/2019 Mixed-Signal SoCs With In Situ Self-Healing Circuitry

    1/13

    Mixed-Signal SoCs With

    In Situ Self-Healing CircuitryChristopher Maxey

    Booz Allen Hamilton

    Sanjay Raman

    Defense Advanced Research Projects Agency

    Kari Groves, Tony Quach, Len Orlando,

    and Aji Mattamana

    Air Force Research Laboratory

    Gregory Creech

    Ohio State University

    Jay Rockway

    SPAWAR Systems Center Pacific

    h ADVANCES IN INTEGRATED circuit technologieshave increasingly enabled the single-chip integra-

    tion of multiple analog and digital functions, result-

    ing in the development of complex mixed-signal

    systems-on-a-chip (SoCs). In particular, scaling

    channel lengths in CMOS processes to 32 nm and

    below has pushed the fmax and ft of silicon CMOS

    transistors in excess of 230 and 440 GHz, respec-

    tively [1] and the fmax and ft of SiGe BiCMOS transis-

    tors in excess of 350 and 300 GHz, respectively [2].

    In this regime, it is possible to realize state-of-the-art

    RF performance on the same platform used to fa-

    bricate digital processors and dense memory cores.

    On the other hand, pattern variations partly due to

    the complex masks needed for subwavelength litho-

    graphy and random variables such as dopant loca-

    tion effects become statistically significant at these

    dimensions. Therefore, a

    major consequence of the

    drive towards ever smaller

    transistor gate lengths is a

    dramatic increase in intra-

    wafer and intradie process

    variability (as measured by

    standard deviation) in crit-

    ical device p aramet ers

    such as threshold voltage,

    effective channel length, etc. Figure 1 plots intra-wafer variability for several process nodes based on

    lithographic limitations published by the ITRS [3]

    along with projections of yield reduction in these

    nodes from Monte Carlo simulations corroborated

    by measurements of representative circuits [4]. In

    aggregate, variability in several parameters simulta-

    neously can greatly exacerbate circuit performance

    degradation. Such variation can have significant

    impact on digital circuits, but the effect is magni-

    fied for analog and mixed-signal circuits due to the

    heightened sensitivity of such designs to device

    mismatch and the integration of numerous individ-

    ual subblocks that can vary greatly in noise charac-

    teristics, operating frequency, etc. Consequently,

    mixed-signal circuit performance yield degrades

    more rapidly with technology scaling than with

    digital circuits.

    To deal with the impact of variability in advanced

    technologies, designers often must conservatively

    accommodate worst-case corner case simulations

    and relax target performance specifications to

    Editors notes:

    This article discusses the goals and recent achievements of the HEALICs

    program. The program_s aim is to enhance wireless systems with sensors,

    actuators, and mixed-signal control loops in order to improve their

    performance yield.

    VHaralampos-G. Stratigopoulos, TIMA Laboratory, and

    Alberto Valdes-Garcia, IBM T. J. Watson Research Center

    0740-7475/12/$31.00 B 2012 IEEENovember/December 2012 Copublished by the IEEE CEDA, IEEE CASS, IEEE SSCS, and TTTC 27

    Digital Object Identifier 10.1109/MDT.2012.2226014

    Date of publication: 23 October 2012; date of current version:

    17 January 2013.

  • 7/28/2019 Mixed-Signal SoCs With In Situ Self-Healing Circuitry

    2/13

    guarantee a sufficient postfabrication parametric

    yield. These traditional corner-based design techni-

    ques require coverage of an extensive and expo-

    nentially growing parameter space that becomes

    intractable for large designs. Furthermore, efforts tocontrol variability through aggressive fabrication

    process management or extensive Design-for-Yield

    (DFY) procedures are costly and ultimately limited

    in effectiveness [5]. An alternative approach advo-

    cated in the Self-HEALing Mixed Signal Integrated

    Circuits (HEALICs) program initiated by the

    authors is to recover lost performance by employ-

    in g on -ch ip sen sors ( meters) , act uat ors

    (knobs) and analog/digital control loops that

    measure the effects of device variability in situ and

    consequently adjust tunable circuit parameters todrive the chip towards a more optimal performance

    point. This control circuitry can be applied at both

    the sub-block and system level allowing the designer

    to focus on performance goals and not on yield

    related issues. As with any control system, chal-

    lenges related to stability, response times, system

    bandwidth, etc. cannot be overlooked when im-

    plementing self-healing, and overcoming these

    challenges while maintaining performance has

    been an important aspect of the program. Further-

    more, HEALICs design teams must also address the

    possibility that process variability adversely im-

    pacts the self-healing circuitry itself, and compen-

    sation for this effect has been a particular point of

    emphasis.Calibration techniques for mixed-signal circuits

    such as analog-to-digital converters (ADCs) have

    been studied extensively; however, they typically

    focus on fully digital techniques that rely on

    known-good states stored in on-chip memory [6].

    Furthermore, analog calibration techniques have

    been investigated for individual circuit blocks [7],

    [8], but they have not been studied extensively for

    full SoC applications. HEALICs technology com-

    bines analog and digital techniques for low-power

    overhead, embedded comprehensive healing forSoCs that compensates not only for process

    variation, but also for the extreme environments

    and long operational lifetimes experienced by the

    Department of Defense (DoD) electronic systems,

    as well as other systems: aviation, automotive,

    industrial automation, etc. For this reason, we

    anticipate that the technology developed under

    this program will greatly enhance long-term

    reliability in addition to improving performance

    yield.

    Figure 1. Variability in critical transistor parameters for submicron process nodes

    plotted along with the associated impact on performance yield for digital and

    mixed signal circuits [3], [4].

    IEEE Design & Test of Computers28

    Digitally Enhanced Wireless Transceivers

  • 7/28/2019 Mixed-Signal SoCs With In Situ Self-Healing Circuitry

    3/13

    Program objectives and planTo measure the efficacy and power efficiency of

    self-healing, the program design teams have targeted

    the yield-improvement and power consumptionoverhead goals described in Table 1.

    Performance yield is defined as the percentage of

    die per wafer that meets or exceeds a set of prede-

    fined circuit performance metrics. The performance

    yield for the baseline (or nonself-healing) die is

    calculated using the following formula:

    Baseline Performance Yield DBaseline

    NBaseline(1)

    where DBaseline is the number of baseline die that

    meet or exceed the same set of performance metricsand NBaseline is the maximum number of baseline die

    that can be fabricated on a single wafer. Without self-

    healing, baseline die are expected to yield poorly if

    held to the same performance standards as the self-

    healing chips. In most cases, the defined perfor-

    mance criteria exceed state-of-the-art for each

    design. In Phase I, the program targeted mixed-signal

    cores or sub-blocks, and in Phase II the program

    targets larger system-on-a-chip designs.

    Posthealing performance yield can be calculated

    using:

    Posthealing Performance Yield DHEALICs

    NBaseline(2)

    where DHEALICs is the number of self-healing die per

    wafer that meet or exceed all metrics. By dividing

    DHEALICs by NBaseline, the performance yield is subject

    to an area overhead correction to ensure that the

    portions of the chip that are exclusively for self-

    healing do not consume an excessive fraction of the

    die area. Combining (1) and (2) yields the following

    formula used for calculating posthealing performance

    yield and accounting for self-healing area overhead:

    Posthealing Performance Yield

    %PosthealingYieldNHEALICs

    NBaseline(3)

    where NHEALICs is the number of self-healing die that

    can be fabricated on a single wafer and %Postheal-

    ingYield is the percentage of self-healing die that

    meet all defined metrics. In some cases, self-healing

    can enable a smaller chip to meet the performance

    metrics compared to a baseline design, i.e.,

    NHEALICs > NBaseline. In this case, (3) allows for the

    performance yield to exceed 100%, which can be

    interpreted as reducing the fabrication costs for agiven part. While not an explicit program require-

    ment, we also expect that self-healing will result in a

    reduction of performance variance as measured over

    the full set of die, indicating that techniques are truly

    compensating for process variability and not for poor

    initial design or inaccurate device models. This effect

    will be readily apparent in the yield measurements

    reported in the following section of this paper.

    Self-healing designs are also required to incur

    minimal power consumption overhead relative to

    the baseline design. Power overhead may be mea-

    sured at an appropriate healing duty cycle if self-

    healing is not expected to be continuously operating.

    While the extreme scaling of silicon transistors is

    primarily responsible for the variability described as

    the motivating factor for self-healing technology, it

    simultaneously makes it possible to incorporate

    substantial processing capabilities on-chip for self-

    healing at low area and power overhead.

    To date, the HEALICs program has demonstrated

    advancements in self-healing yield and performance

    Table 1 HEALICs program goals for self-healing.

    November/December 2012 29

  • 7/28/2019 Mixed-Signal SoCs With In Situ Self-Healing Circuitry

    4/13

    in RF transceivers and subblocks, including ADCs,

    phase-locked loops (PLLs), and power amplifiers

    (PAs). These designs have been fabricated in

    processes ranging from 180 nm SiGe BiCMOS to

    32 nm silicon-on-insulator (SOI) CMOS, demonstrat-

    ing the efficacy of self-healing across a range oftechnologies and variability effects. Table 2 sum-

    marizes the individual program design teams and

    the mixed-signal SoCs they have targeted for

    demonstrating their self-healing methods.

    This paper will describe results obtained by the

    IBM, Raytheon, and UCLA teams, however, several

    other important self-healing research activities in this

    program are not covered in detail because of space

    constraints. Examples include self-healing image

    reject mixers [9] and LNAs [10] developed at

    Georgia Tech (a member of the NGES team), novelI-gate MOSFETs for self-healing circuits [11] devel-

    oped at University of Texas at Dallas (a member of

    the IBM team), and built-in self-test synthesis tech-

    niques for self-healing [12] developed at Auburn

    University (a member of the BAE Systems team).

    Recent accomplishmentsThis section will cover the measurement and

    yield improvement results for several of the fabri-

    cated and measured Phase I designs.

    Self-healing PA and PLL designs

    A research team led by IBM has demonstrated a

    self-healing 28 GHz PA in 45 nm SOI and a self-

    healing 2127 GHz PLL in 32 nm SOI. Together, these

    components were designed for a Ka-band radio

    transceiver.The high-level schematic of the self-healing PA

    designed by CalTech is shown in Figure 2 [13]. The on-

    chip integrated sensors include RF power sensors at

    the input and output to estimate gain, junction

    temperature sensors to assist with the estimation of

    power added efficiency (PAE), and dc current sensors

    to determinepower consumption. Thepower sensor is

    calibrated by measuring the output of a few represen-

    tative die and calculating a ratio between sensed

    power and actual power. These data are included in

    the healing algorithm for each of the chips. On-chipcontrol knobs include bias actuation mechanisms at

    each of the PA stages and tunable transmission lines at

    the PA output used to adjust matching impedances.

    An embedded fully custom microprocessor is used to

    determine the knob settings necessary to minimize

    power consumption for a desired output power (thus

    maximizing PAE). The control algorithm affords a

    level of flexibility in operation and the target output

    power can be designated at run time. Through this

    control loop, the PAE is increased from 5.6% in the

    Table 2 Summary of target self-healing demonstration circuits.

    IEEE Design & Test of Computers30

    Digitally Enhanced Wireless Transceivers

  • 7/28/2019 Mixed-Signal SoCs With In Situ Self-Healing Circuitry

    5/13

    baseline case to 7.3% in the healed state. PAE is

    defined as:

    PAE100RF Output PowerRF Input Power

    dc power consumption:(4)

    Furthermore, the 1 decibel compression point

    and the gain are increased from 11.3 to 13.8 dBm

    and from 20.3 to 23.7 dB, respectively.

    This PA chip was also an excellent platform for

    investigating the potential efficacy of self-healing for

    Figure 2. Self-healing PA: (a) high-level schematic and (b) die photograph (1.8 mm 0 1.7 mm).

    November/December 2012 31

  • 7/28/2019 Mixed-Signal SoCs With In Situ Self-Healing Circuitry

    6/13

    improving RF/mixed-signal circuit reliability. To this

    end, CalTech performed an extreme test of reliability

    healing by intentionally and sequentially inducing

    damage in one of the PA output stages through theuse of laser ablation and initiating the healing

    algorithm after each stage of damage. Notably,

    self-healing was capable of recovering significant

    amounts of lost output power by autonomously ad-

    justing the output matching to compensate for the

    change in impedance loading. A set of curves show-

    ing output power data for the undamaged and for

    three successive ablations is shown in Figure 3 along

    with before and after photos of the PA. Ultimately,

    the self-healing algorithm is shown to recover up to

    4.8 dBm of output power compared to the baselinesettings.

    Concurrent with the design of the PA, the IBM

    team designed and tested a 25-GHz self-healing

    PLL fabricated in 32 nm SOI technology. The PLL

    architecture was a hybrid analog/digital design

    with proportional and integral paths following the

    phase/frequency detector. A peak detector at the

    output of the oscillator acts as the primary sensing

    path for the self-healing algorithm, which, in this

    case, is primarily responsible for reducing phase

    noise at a 10-MHz offset from the carrier to

    G124 dBc/Hz. The bias values and VCC are

    measured along with the frequency to complete

    the in situ virtual phase noise model of the PLLperformance [14]. During testing, a quadratic

    relationship is derived that relates the output

    amplitude, output frequency, bias voltage, bias

    current, tuning band, and VCC value to the mea-

    sured phase noise. Once this equation is deter-

    mined, the coefficients for each term are stored in

    on-chip memory. After self-healing is initiated, the

    knob settings are stepped through incrementally. An

    on-chip processor then evaluates the stored equa-

    tion for the sampled sensor values and returns an

    estimate of the phase noise. If phase noise is at aminimum, the algorithm stops and the knob settings

    are held constant. The algorithm is implemented

    completely on-chip in a 16-bit arithmetic logic unit

    which uses a 32-word data memory and a 256-word

    instruction memory for low healing overhead. Using

    this algorithm, the performance yield of the PLL is

    improved from $20% in the baseline case to 100%

    posthealing for 55 measured die. Healing power

    overhead was $10% and healing area overhead

    was 12.7%.

    Figure 3. Self-healing result for intentionally damaged PA: (a) no damage; (b) 1/8 stage removed;

    (c) 1/4 stage removed; and (d) 1/2 of stage removed. Micrographs of PA: (e) before and

    (f) after damage.

    IEEE Design & Test of Computers32

    Digitally Enhanced Wireless Transceivers

  • 7/28/2019 Mixed-Signal SoCs With In Situ Self-Healing Circuitry

    7/13

    Self-healing pipelined ADC

    R esearchers f rom UC LA

    have pursued the design of a

    self-healing ultralow-power pi-

    pelined 1 Gsps ADC with a flash

    output stage in 65 nm CMOS

    [15]. Therefore, canonical self-healing involved adding control

    loops and processors to the

    baseline chip with the con-

    straint of minimizing the over-

    head created by these auxiliary

    blocks. For this ADC design,

    however, the application of

    self-healing was found to signif-

    icantly decrease the size of the

    yielding part relative to a non-

    yielding baseline part by en-abling the use of minimum

    sized devices in critical analog

    parts of the design.

    Flash ADCs are well known

    for realizing extremely high

    sample rate data converters at

    the expense of high comparator

    cou nt f or h igh n umbers of

    digital output bits. Further-

    more, these comparators ex-

    hibit significant capacitanceand gain mismatch in deeply

    scaled CMOS that would other-

    wise be an attractive fabrication

    process for such ADCs. One

    method to combat the mis-

    match (the baseline approach)

    is to use devices with larger

    than minimum dimensions that

    exhibit lower variability stan-

    dard deviation; but this ob-

    viates the speed, the size, andthe power advantages of using

    65 nm or better technology.

    Alternatively, a self-healing al-

    gorithm designed to mitigate

    the effect of comparator mis-

    match would enable the use of

    minimum size devices. Specifically, in this imple-

    mentation, capacitance and gain mismatch healing

    is employed in the 4-bit input stage and capaci-

    tance mismatch healing is employed in the follow-

    ing four stages. The final three stages have sufficient

    performance margin to not require additional

    healing. Figure 4(a) shows the general ADC

    architecture schematic.

    Figure 4. (a) High-level ADC schematic showing the self-healing required

    for each stage. Inset: op-amp comparator schematic and (b) comparison

    of self-healing design to recently published ADCs in various submicron

    CMOS technologies. (All data points except for self-healing points

    compiled from [16]).

    November/December 2012 33

  • 7/28/2019 Mixed-Signal SoCs With In Situ Self-Healing Circuitry

    8/13

    Through this approach, the ADC demonstrated

    56.7 dB of signal-to-noise-and-distortion ratio

    (SNDR) equivalent to 8.38 effective bits with a

    491 MHz input signal sampled at 1 Gsps. More

    importantly, the power consumption was merely

    33 mW, resulting in an ADC figure-of-merit (FOM)

    of$99 fJ/conversion step. The ADC FOM is calcu-lated as follows:

    FOM Power

    2ENOB Sampling Frequency: (5)

    Compared to a baseline design utilizing larger

    comparator transistors, this design was 41 smaller

    and $100 less power consumptive. A conservative

    clock tree design consumed approximately a third of

    the power and could have been further reduced

    with aggressive design. Furthermore, simulation re-

    sults indicate approximately a 16% reduction in theFOM if the design was ported to 45 nm technology. A

    plot comparing this result to others in recent liter-

    ature is shown in Figure 4(b).

    Self-healing radar receivers

    In addition to subblocks dedicated for use in

    communications systems, self-healing has been

    shown to dramatically increase performance yield

    for DoD-relevant radar and EW receivers as well. For

    example, a team led by Raytheon developed a

    618 GHz self-healing radar receiver with several on-chip sensors and actuators designed to reduce I/Q

    phase mismatch, PLL output spur level, and gain

    fluctuations across the spectral field of regard [17].

    Figure 5a shows the design schematic.

    Unique to this team, an on-board ARM process-

    ing core was synthesized and fabricated to run a

    modified Nelder-Mead optimization algorithm used

    to efficiently minimize mismatch, suppress spurs

    and flatten gain response. The ARM core utilizes

    130.4 KB of memory, but only 36 KB are specific to

    the healing circuitry. In this case, HEALICs techni-

    ques are shown to take advantage of baseline pro-

    cessing and memory capabilities to reduce healing

    area and power overhead. To reduce phase mis-

    match, for example, an auxiliary mixer is used to

    produce a dc value proportional to the amount of

    phase offset between the Iand Q paths. Feedback to

    variable amplifiers in each path are used to fine tune

    the phase offset, and the Nelder-Mead algorithm is

    used to establish values for each amplifier that mini-

    mizes mismatch. The search routine is applicable to

    bound-constrained and discretized optimization

    problems and has shown efficacy for the types of

    self-healing required for the radar receiver designed

    in this effort. Furthermore, Nelder-Mead is particu-

    larly robust in the presence of noisy measurement

    data, which is expected from the low-power sensors

    typically incorporated in self-healing designs. Toheal PLL spur levels, a similar algorithm is used with

    slightly different knobs and sensors. A power de-

    tector on the output is fed to the algorithm, which

    independently sets two knobs: charge injection on

    the VCO control line and a trigger setting to establish

    the precise timing of the charge injection. Finally, to

    heal gain flatness, the output power sensor is mo-

    nitored and the tunable amplifiers are further ad-

    justed while maintaining optimum phase matching.

    Figure 5b shows histograms for healed mismatch

    and gain flatness. The area shaded in green repre-sents the performer-defined metric that sets the stan-

    dard for yield in the case of this design. For both of

    these metrics, yield was improved from 0% in the

    baseline case to 100% in the healed case.

    The Raytheon team is also investigating the

    effects of ageing on circuits, which is of particular

    interest given that some defense, as well as com-

    mercial, electronic systems are fielded for decades.

    The main cause of aging in PMOS is negative bias

    temperature instability (NBTI) which shifts the

    threshold voltage over time and is especially delete-rious for gate lengths shorter than 100 nm. Hot-

    carrier injection (HCI) is an important in ageing

    mechanism for NMOS and has an amplified impact

    on mixed signal circuits compared to pure digital

    circuits. NCSU, a member of the Raytheon team, is

    developing techniques for the purposes of designing

    self-healing knobs and sensors specifically for

    ageing and has published initial results in [18].

    Self-healing 60 GHz radio transceiver

    Another research group at UCLA has successfully

    incorporated self-healing sensors, actuators and

    control loops to substantially improve the perfor-

    mance yield of a 60 GHz, 4 Gbps communications

    transceiver fabricated in 65 nm CMOS [19]. The

    schematic for this design, identifying additions spe-

    cific to self-healing, and the photograph of the fabri-

    cated die are shown in Figure 6.

    Circuitry dedicated solely to self-healing repre-

    sented 8% of the total area. Likewise, the power

    consumption of these blocks amounted to only 3.7%

    IEEE Design & Test of Computers34

    Digitally Enhanced Wireless Transceivers

  • 7/28/2019 Mixed-Signal SoCs With In Situ Self-Healing Circuitry

    9/13

    of the chips total. On-chip sensors include an envel-

    ope sensor for tracking the output spectrum; a tem-

    perature sensor for estimating the kT noise level; apower sensor for estimating the path gain; and an

    auxiliary ADC for quantizing the sensor outputs. Self-

    healing control knobs include phase offset and dc

    offset for the individual I/Q paths; PA and mixer

    current biases; variable gain LNAs; and an auxiliary

    DAC for generating control signals. The transceiver

    utilizes a loop-back healing approach, where the Tx

    is used to drive the Rx for the purposes of measuring

    the radios performance. This minimizes the need

    for off-chip calibration of the on-chip sensors. The

    central component for mediating the self-healing

    algorithms is the parameter estimator (PE) block

    shown in the block diagram in Figure 6a. The (PE)comprises a 128-point FFT processor synthesized in

    digital logic along with spectrum magnitude estima-

    tion for both the Iand Q channels. The PE is respon-

    sible for estimating LO leakage power, image tone

    power, among other parameters and consumes only

    0.56 mm2 of die area and 4.8 mW. The DAC control-

    ler, meanwhile, provides test tones for measuring

    circuit response to 1-tone and 2-tone tests.

    Critical performance parameters for this design

    include receiver noise figure; transmitter linearity;

    Figure 5. (a) Radar receiver schematic showing location and connections for self-healing sensors and

    knobs; histograms showing improvement in performance yield for phase error (b) and gain flatness (c).

    November/December 2012 35

  • 7/28/2019 Mixed-Signal SoCs With In Situ Self-Healing Circuitry

    10/13

    Figure 6. 60 GHz, 4 Gbps self-healing transceiver: (a) high-level schematic; (b) die

    photograph (4 mm4 mm); and performance yield enhancement for (c) image tone

    rejection and (d) receiver NF.

    IEEE Design & Test of Computers36

    Digitally Enhanced Wireless Transceivers

  • 7/28/2019 Mixed-Signal SoCs With In Situ Self-Healing Circuitry

    11/13

    and image tone power. The target noise figure was

    6 dB, and self-healing was able to correct the

    baseline value of 9.4 dB to 4.9 dB for 100% of the

    chips. Likewise, the target Tx third-order output

    intermodulation product level (OIM3) and the target

    image tone rejection level were both 40 dBc, and

    the on-chip algorithm was able to correct baselinevalues of32.4 dBc and 32.8 dBc to 42.6 dBc

    and 41.8 dBc, respectively, on 100% of the die.

    To heal the image tone power, the envelope

    sensor captures the spectrum of the transmitter out-

    put and the PE estimates the magnitude of the power

    in each of the FFT frequency bins. After the estima-

    tion is performed, the phase offset, relative gains and

    dc offsets between the I/Q channels are adjusted to

    minimize the power at the image frequency. A

    coarse/fine search algorithm is implemented in cus-

    tom logic to close the control loop and minimize the

    measured image power. Likewise, to heal the noise

    figure, the temperature sensors are sampled by the

    auxiliary ADC to estimate the current kT level. Given

    the relationship between stage gain and overall

    noise figure, the NF can be optimized by adjusting

    the gain in each of the three stages. Figure 6c and d

    show histograms for each of these performance pa-rameters comparing the healed state to the baseline

    state. Since the area of the baseline design is con-

    strained by the number of bond pads and since the

    self-healing circuitry fit in vacant parts of the die,

    there is no area overhead correction necessary for

    this part.

    IN SITU SELF-HEALING has been shown to dramat-

    ically improve the performance yield of RF/mixed-

    signal designs at deep-submicron nodes. A wide

    range of cores and subblocks from PAs to entire

    Table 3 Summary of yield improvement through self-healing for the circuits described in this paper.

    November/December 2012 37

  • 7/28/2019 Mixed-Signal SoCs With In Situ Self-Healing Circuitry

    12/13

    transceivers have been co-integrated with sensors,

    actuators, and control loops, resulting in the resto-

    ration of lost performance at low power and die area

    overhead. A summary of some of the performance

    yield improvement capabilities highlighted in this

    paper is included in Table 3.

    Activities that promise to extend these capabil-ities to the system-on-a-chip level are already

    underway, and a repository of self-healing IP has

    been established at the Air Force Research Labo-

    ratory to facilitate a more widespread adoption of

    these techniques throughout the DoD design

    community.

    AcknowledgmentsThe authors would like to thank all of the per-

    formers on the HEALICs program for their effortstowards achieving the program goals; in particular,

    those who contributed to the work outlined in this

    paper: principal investigators Dr. Jose Tierno and

    Mr. Daniel Friedman from IBM, Prof. Behzad Razavi

    from UCLA, Mr. Gerry Sollner from Raytheon, Prof.

    Frank Chang also from UCLA, and their teams. The

    HEALICs program was developed at and funded by

    the Defense Advanced Research Projects Agency.

    The work presented in this paper was supported

    through Air Force Contracts FA8650-09-C-7924,

    FA8650-09-C-7925, and FA8650-09-C-7926; Navygrants N66001-09-1-2029 and N66001-09-1-2030; and

    Navy contract N66001-09-C-2023. The views, opi-

    nions, and/or findings contained in this article are

    those of the authors and should not be interpreted

    as representing the official views or policies, either

    expressed or implied, of the Defense Advanced

    Research Projects Agency or the Department of

    Defense. h

    h References

    [1] C.-H. Jan et al. RF CMOS technology scaling in

    high-k/metal gate era for RF SoC (system-on-chip)

    applications, in Proc. 2010 IEEE Int. Electron

    Devices Meeting (IEDM), Dec. 68, 2010,

    pp. 27.2.127.2.4.

    [2] M. Khater et al. SiGe HBT technology with

    fmax=fT 350=300 GHz and gate delay below 3.3 ps,

    in Proc. 2004 IEEE Int. Electron Devices Meeting

    (IEDM), Dec. 1315, 2004, pp. 247250.

    [3] International Technology Roadmap for

    Semiconductors, 2011 Edition, 2011.

    [4] K. Bernstein et al. High-performance CMOS

    variability in the 65-nm regime and beyond, IBM J.

    Res. Dev., vol. 50, no. 4.5, pp. 433449, Jul. 2006.

    [5] J. Hartmann, Towards a new nanoelectronic

    cosmology, in Proc. 2007 IEEE Int. Solid-State

    Circuits Conf. (ISSCC), Feb. 1115, 2007, pp. 3137.

    [6] J. A. McNeill et al. FSplit ADC_ calibration for all-digitalcorrection of time-interleaved ADC errors, IEEE

    Trans. Circuits Systems II: Expr. Briefs, vol. 56, no. 5,

    pp. 344348, May 2009.

    [7] Y. Liu et al. CMOS RF power amplifier variability

    and reliability resilient biasing design and analysis,

    IEEE Trans. Electron Devices, vol. 58, no. 2,

    pp. 540546, Feb. 2011.

    [8] K. Jayaraman et al. Design and analysis of 160 GHz,

    RF CMOS peak detectors for LNA calibration, in

    Proc. 2009 Int. Symp. VLSI Design, Automat. Test,

    Apr. 2830, 2009, pp. 311314.

    [9] P. K. Saha et al. An adaptive, wideband SiGe image

    reject mixer for a self-healing receiver, in Proc. 2011

    IEEE Bipolar/BiCMOS Circuits Technol. Meeting

    (BCTM), Oct. 911, 2011, pp. 99102.

    [10] D. C. Howard et al. A UWB SiGe LNA for multi-band

    applications with self-healing based on DC extraction

    of device characteristics, in Proc. 2011 IEEE

    Bipolar/BiCMOS Circuits Technol. Meeting (BCTM),

    Oct. 911, 2011, pp. 111114.

    [11] C. Wu et al. I-gate body tied silicon-on-insulator

    MOSFETs with improved high frequency

    performance, IEEE Electron Device Lett.,

    vol. 21, no. 1, pp. 4345, Jan. 2011.

    [12] G. J. Starr et al. Automated generation of built-in

    self-test and measurement circuitry for mixed-signal

    circuits and systems, in Proc. 24th IEEE Int. Symp.

    Defect Fault Tolerance in VLSI Syst., Oct. 79, 2009,

    pp. 1119.

    [13] S. M. Bowers et al. A fully-integrated self-healing

    power amplifier, in Proc. 2012 RFIC Symp.,

    Montreal, PQ, Canada, Jun. 2012.

    [14] S. Yaldiz et al. Virtual phase noise sensor for

    self-healing voltage controlled oscillators, in Proc.

    2011 Government Microcircuit Applicat. Critical

    Technol. Conf., Mar. 2011.

    [15] B. D. Sahoo et al. A 10-Bit 1-GHz 33-mW CMOS

    ADC, in Proc. 2012 IEEE Symp. VLSI Technol.,

    Jun. 2012.

    [16] J. Mulder et al. An 800 MS/s dual-residue pipeline

    ADC in 40 nm CMOS, in Proc. 2011 IEEE Int.

    Solid-State Circuits Conf., Feb. 1923, 2011,

    pp. 184186.

    IEEE Design & Test of Computers38

    Digitally Enhanced Wireless Transceivers

  • 7/28/2019 Mixed-Signal SoCs With In Situ Self-Healing Circuitry

    13/13

    [17] G. Sollner et al. Tunable receiver for 618 GHz

    with autonomous self-healing, in Proc. 2011

    Government Microcircuit Applicat. Critical Technol.

    Conf., Mar. 2011, pp. 4952.

    [18] M. B. Yelten et al. Analog Negative Bias Temperature

    Instability (NBTI) monitoring circuit, IEEE Trans.

    Device Mater. Reliab., vol. 12, no. 1, pp. 177179,Mar. 2012.

    [19] A. Tang et al. A low overhead self-healing embedded

    system for ensuring high performance yield and

    long-term sustainability of a 60 GHz 4 Gbps

    radio-on-a-chip, in Proc. 2012 IEEE Int. Solid-State

    Circuits Conf. (ISSCC), Feb. 1923, 2012.

    Christopher Maxey is an Associate with Booz

    Allen Hamilton. His research interests include mixed-

    signal and RF circuits. He has a Masters degree in

    electrical engineering from Virginia PolytechnicInstitute and State University (Virginia Tech) in

    2004. He is a member of the IEEE.

    Sanjay Raman is a Program Manager at DARPA,

    currently on assignment from Virginia Polytechnic

    Institute and State University (Virginia Tech), where

    he is a Professor of Electrical and Computer

    Engineering. His research interests include silicon-

    based RF/microwave millimeter-wave circuits; self-

    correcting RF/mixed-signal circuits; heterogeneous

    integration; and sensor microsystems. He has a PhD

    in electrical engineering from the University of

    Michigan in 1998. He is a senior member of the IEEE.

    Kari Groves is an RF Engineer with the Air Force

    Research Laboratory.

    Tony Quach is an RF Engineer with the Air Force

    Research Laboratory.

    Len Orlando is an RF Engineer with the Air Force

    Research Laboratory.

    Aji Mattamana is an RF Engineer with the Air

    Force Research Laboratory.

    Gregory Creech is recently retired from the AirForce Research Laboratory. He is now with Ohio

    State University.

    Jay Rockway is with SPAWAR Systems Center

    Pacific.

    h Direct questions and comments about this article

    to Chri stopher M axey , B ooz A ll en Hami lton;

    [email protected].

    November/December 2012 39