Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS 1

A Statistical Design Approach Using Fixed


and Variable Width Transconductors for
Positive-Feedback Gain-Enhancement OTAs
Andrew Peter Nicholson, Astria Nur Irfansyah, Julian Jenkins, Tara Julia Hamilton, Member, IEEE,
and Torsten Lehmann, Senior Member, IEEE

Abstract— The positive-feedback gain-enhancement opera- required to maintain low area (high unity gain frequency), low
tional transconductance amplifier (OTA) design is a promising power and high gain [4], [5]. Digitally assisted analog circuits
architecture to scale into deep submicron CMOS. Ever smaller are a solution for active mismatch cancelation, however,
CMOS process nodes require analog circuit designs that can
overcome the area–power-matching relation. We introduce a this introduces new challenges, such as the classification of
Nauta OTA with a split architecture consisting of fixed width and errors (time dependence and monotonicity) and the design of
digitally programmable variable width transconductors utilizing calibration and linearization techniques [6], [7].
the minimum grid-spacing of the CMOS process enabling an To design operational transconductance amplifiers (OTAs)
active mismatch cancelation technique. A variation-aware sta- further into deep submicron CMOS using positive-feedback
tistical design practice is introduced to analyze the sizing of
transconductors, computing code-word solutions for statistically gain-enhancement techniques that rely upon device matching,
likely solutions, and estimating average maximum dc gain over we need to proceed with a mismatch tolerant architecture.
the entire code-space of many simulated OTAs. Prototypes of The Nauta OTA is a high-speed single-stage design, which
a 8-bit differential OTA in 180-nm CMOS designed using the is suitable for digitally assisted variation-aware techniques
Nauta structure fixed width and digitally programmable variable to scale down into deep submicron CMOS [8]–[10]. In
width architecture achieves an average maximum dc gain of
60 dB, simulated unity gain frequency of 4.6 GHz, and a this paper, we introduce a circuit architecture and statistical
figure-of-merit of 1 GHz/mW. design procedure for digitally assisted positive-feedback gain-
enhancement OTAs. The OTA is split into fixed width and
Index Terms— Deep submicron CMOS, digitally assisted
analog design, positive-feedback gain-enhancement operational digitally programmable variable width components using a
transconductance amplifier (OTA). statistical approach of designing the sizes of each transconduc-
tor for high-gain solutions. A Monte Carlo-based simulation
method of quantifying our various yields and average dc gains
I. I NTRODUCTION is outlined, and then compared with the experimental results
of prototypes fabricated in 180-nm CMOS.
A NALOG designers are faced with numerous challenges
to scaling down circuit designs into deep submicron
CMOS—increasing output conductance of short-channel
Our motivation for designing with a statistical frame-
work for a mismatch tolerant architecture is that the gain-
devices implies decreasing intrinsic gain, low voltage enhancement technique used in the Nauta OTA architecture
headroom as voltage supply decreases faster than threshold relies upon device matching. However, in deep submicron
voltages, and increased nonlinearities [1]–[3]. In 90 nm and CMOS, minimum-sized device matching is becoming statis-
below, the traditional analog designs will not be able to break tically worse, for instance, as measured by comparing the
the area–power-matching relation, and simply scaling up width standard deviation of drain currents between two minimum
and length to overcome mismatch is limited by gate leakage sized identical devices within the same die [11], [12]. Our
effects. Thus, active mismatch cancelation techniques are design presented here is informed by statistical techniques
to enhance mismatch tolerance through the availability of a
Manuscript received July 25, 2016; revised November 30, 2016; accepted series of transconductance values implemented as digitally
January 3, 2017. This work was supported by the Australian Research Council
Industry Linkage under Grant LP100200275. programmable analog components whose width differences are
A. P. Nicholson and T. Lehmann are with the School of Electrical defined by a function of the minimum CMOS grid-spacing.
Engineering and Telecommunications, University of New South Wales, Recent work on mismatch tolerant analog design includes
Sydney, NSW 2033, Australia.
A. N. Irfansyah is with the School of Electrical Engineering and Maunu et al. [13] who present a design of a self-calibrating
Telecommunications, University of New South Wales, Sydney, NSW 2033, current source utilizing the inherent variations within an array
Australia, and also with the Department of Electrical Engineering, Institut of minimum-sized devices. Ragab et al. [14] introduce a
Teknologi Sepuluh Nopember, Surabaya 60117, Indonesia.
J. Jenkins is with Perceptia Devices Inc., Sydney, NSW 60117, Australia. component redundancy and random diversity architecture to
T. J. Hamilton is with The MARCS Institute, Western Sydney University, harness randomness to calibrate a positive-feedback amplifier
Penrith, NSW 2150, Australia. utilizing minimum-sized devices biased in subthreshold.
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. This paper will detail the statistical design process showing
Digital Object Identifier 10.1109/TVLSI.2017.2657885 how we can partition the transconductors into fixed width and
1063-8210 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS

symmetrical architecture for both paths in the differential


circuit, we can find the result of ideal width difference
between self-coupled and cross-coupled transconductors by
considering just three transconductors in one path and an
assumption of a constant ratio of transconductance to output
conductance operating point [8], [16].
In Sections III-A–III-D, we define a code-word as the setting
of the digital control lines for the digital OTA, refer to Fig. 2.
P is the control line to power-up the OTA. P, E N[7 : 0] is
in bus notation, indicating a 9-bit bus composed of powerup
control line for the fixed width self-coupled transconductor
and the 8-bit variable width transconductor programming bus.
As a simplifying assumption in this paper, we assume that
the code-word is symmetrical across both the paths, and the
active low control lines are always the inverse of the active
Fig. 1. Nauta differential OTA schematic. high lines. This means that the total size of the code-space
is 2 N for a Nauta differential OTA using N-bit digital-to-
variable width components, size the individual components of transconductance converters [17].
the variable width transconductors for the highest likelihood Our architecture for the digital Nauta OTA that follows is an
of combinatorial matching, and estimate the statistical average improvement over full programmability of transconductance
of the maximum dc gain achievable for the digital Nauta OTA. in all components [8] with a computationally feasible size of
It can achieve higher gains than traditional analog multistage the code-space. We have only fixed-width components for the
OTA designs whilst using a comparable silicon die area but feedforward and cross-coupled transconductors, and for the
achieves higher unity gain frequencies. In terms of a figure-of- self-coupled transconductors, we use a parallel fixed width and
merit (FoM) as measured in a unity gain frequency per power variable width transconductor design. This new architecture
consumption ratio (GHz per mW) our designs are competitive provides a small code-space that is easily traversable by a
to the state-of-the-art. Further into deep submicron CMOS, calibration algorithm combined with a statistical technique of
the digital Nauta OTA increases unity gain bandwidth and choosing design parameters based on a grid-spacing design of
decreases silicon area whilst still delivering high dc gains. digitally controllable transconductor width.
This paper is organized as follows. Section II is the We acknowledge the inherent device mismatch in the design
mathematical analysis of the design process for the fixed of the programmable width transconductor architecture by
width and variable width components of the positive-feedback using a certain number of tristate-able parallel connected
gain-enhancement OTA and expectation of calibrated dc gain, minimum-length transconductors of varying widths. Starting
Section III is a model of the unity gain frequency of the OTA, with a minimum width transconductor, the sequence is defined
Section IV is a statistical analysis over the entire code-space, relative to the previously defined device width plus a width
which can quantify the performance metrics of the OTA, increment that is a multiple of the grid-spacing resolution
Section V is a description of the calibration strategy of the of the CMOS process. In this way, we combine the idea
digital Nauta OTA, Section VI is the experimental results of redundancy—using multiple transconductors varying in
obtained from ten prototype digital OTAs in 180-nm CMOS, width by only by a single grid-spacing resolution—with the
Section VII is a comparison with a traditional analog amplifier statistical benefits produced by a combinatorial space of the
design, and Section VIII is the conclusions of the research. digitally programmable transconductors with width differences
at integer multiples of the grid-spacing resolution. We define
the parameter M as the number of devices in the series starting
II. S IZING S TRATEGY OF T RANSCONDUCTORS from a minimum length, minimum width device that are sized
IN D IGITAL NAUTA OTA
apart in width only by a single CMOS grid-spacing. The
The width sizing strategy for transconductors is based on parameter N is the total number of independent devices in
the method of gain enhancement used in the Nauta OTA [15]. the series, inclusive of M.
This is achieved using positive feedback to form a negative
transconductance, which cancels the output conductance of the
circuit and produces large gain. Gain is therefore restricted to A. Fixed Width Feedforward and
transconductance (or width) matching resolution. We assume Cross-Coupled Transconductors
minimum length for all transistors in our structure as we are The feedforward fixed width component is derived from
targeting the highest unity gain frequency possible. the transconductance requirements, gm , and dc gain speci-
Referring to Fig. 1, the cross-coupled (Inv3, Inv6) fication, A V . We illustrate our argument with an example
transconductors’ width should be larger than the self-coupled specifications of 4-mS transconductance minimum, gm,spec,
(Inv4, Inv5) transconductors’ widths, to form the negative (relative to a single-ended transconductor) and at least a linear
transconductance, and this should be equal to the sum of dc gain of 64 (36 dB) that can be used within a sigma–delta
output conductance of all transconductors. Assuming a modulator [10].
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

NICHOLSON et al.: STATISTICAL DESIGN APPROACH USING FIXED AND VARIABLE WIDTH TRANSCONDUCTORS 3

Fig. 2. Digital Nauta differential OTA using two 8-bit digital-to-transconductance converters for the two variable width components of the self-coupled
transconductors (Inv4, Inv5). This OTA has values N = 8, M = 2. P is the control line to power-up the OTA. P, E N [7 : 0] is in bus notation, indicating a
9-bit bus composed of power-up control line for the fixed width self-coupled transconductor and the 8-bit variable width transconductor programming bus.

To a first order, the width matching considerations require We have


that in a digitally programmable Nauta structure, the feed-
forward transconductor’s width is at least half the linear dc WCC = k · WFF 0 < k < 1. (2)
gain as large as the smallest width transconductor, Wmin [8]. The self-coupled transconductor’s width can be determined
This accounts for the first estimate in (1). This lower bound by the necessity of setting the transconductance difference
on feedforward transconductor width, WFF , can then be con- between self-coupled and cross-coupled transconductors equal
firmed to match gm,spec specifications through transconduc- to the sum of output conductance of all transconductors for
tance measurements across process variation-only Monte Carlo maximum dc gain. Thus
simulations (using 4096 trials) to ensure that it is greater than
the gm,spec specification to within 3σ . E[] is the expectation |(WSC − WCC ) · gm | = (WSC + WCC + WFF ) · gds. (3)
operator Following from (3), we solve for the width of the self-
1 coupled transconductors for the condition of maximum dc
WFF = · E[ A V ] · Wmin . (1) gain as:
2
Next, this transconductor is statistical sampled via process WSC · (gm + gds) = WCC · (gm − gds) − WFF · gds
variation and mismatch Monte Carlo trials to find the mean WSC = (ζ − ξ ) · WCC − ξ · WFF (4)
and standard deviation of transconductance, gm , and output
conductance, gds . These parameters will be used to estimate where ζ −1 = 1 + μ(gds/gm ) and ξ = ζ · μ(gds/gm ). We
the range of width differences between self- and cross-coupled define μ(gds/gm ) as the mean of the ratio of the output
transconductance needed to establish maximum dc gain conductance to transconductance.
conditions. Substituting (2) into (4) allows us to solve for the ideal self-
The cross-coupled fixed width, WCC , component is defined coupled width (in the absence of device mismatch) solely in
as a fraction, k, of WFF . Total self-coupled transconductor terms of cross-coupled width to establish maximum dc gain.
width (fixed and variable components) is defined as WSC . We have
A tradeoff for the selection of this fraction exists between min- WSC = (ζ − (k −1 + 1) · ξ ) · WCC . (5)
imizing for power, area, and hysteresis effects (smaller k) and
the requirement of the common mode stability condition [18], Defining W as the ideal width difference between
that is, WSC + WCC > WFF (larger k). cross-coupled and self-coupled transconductors for maximum
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS

TABLE I the overall ideal width. For a first approximation, the fixed
F IXED C OMPONENTS S UMMARY FOR D IGITAL N AUTA OTA width component of the self-coupled transconductors can be
estimated by subtracting a value greater than W from WCC .
We cannot determine a satisfactory value exactly until after
we also choose sizes for the variable components and conduct
parametric evaluations of the entire OTA and simulate the
achievable parametric yields.
The variable component for the self-coupled transconduc-
tors, WSC,variable , is composed of a set of N parallel-connected
tristate transconductors varying in width. Each transconductor
is independently digitally controllable via the tristate switches,
which are sized the same as the corresponding signal carrying
transistors.
Given a particular programming code-word, C, for the vari-
able width self-coupled transconductors, 0 ≤ C ≤ 2 N − 1, in
combination with the fixed width self-coupled transconductor,
for the maximum dc gain condition, we have

N
dc gain, and using (5), we find WSC,fixed + WSC,variable,i · F(C, i ) = WSC (9)
W = WCC − WSC i=1

W = WCC · (1 − (ζ − (k −1 + 1) · ξ )). (6) where F(C, i ) is a function, which is one only if the i th digit
of the binary form of the code-word C is one, otherwise zero.
From (2) and (5), we can determine if the common mode As (9) reveals that the value we ultimately chose for
stability condition is met, that is, if WSC + WCC > WFF holds WSC,fixed will be a tradeoff between having a larger fixed com-
at the condition of maximum dc gain ponent to minimize mismatch error in the fixed components
and fewer N bits needed to cover a smaller width range—
WSC > WFF − WCC
versus a smaller fixed component and larger width range,
WSC > WCC · (k −1 − 1) allowing increased coverage of grid-spaced width solutions at
ζ − (k −1 + 1) · ξ > k −1 − 1. (7) the expense of more N bits.
Given an N-bit implementation of the variable width self-
We start our numerical calculations using the smallest width
coupled transconductors there are N transconductors with the
transconductor available in the 180-nm CMOS process—a
width of the xth device defined via a first-order difference
220 nm:180 nm (W : L) nMOS device with an nMOS:pMOS
between successive transconductors’ width, as so
ratio of 1:4. Using Wmin = 220 nm and E[ A V ] = 64 with (1),
we obtain WSC,variable,x − WSC,variable,x−1 = f (x) ·  for 1 < x ≤ N
64 (10)
WFF = · 220 nm = 7.04 μm. (8)
2
where  is the grid-spacing—the minimum increment for
Substituting this result for WFF and using k = (5/7) into (2), width in the CMOS process.
we obtain WCC = 5.03 μm. Finally, using the values for The discrete-valued function f (x) is defined for 1 < x ≤ N,
the mean ratio of transconductance to output conductance given a programmable transconductor architecture with N bits
and values for ζ, ξ as shown in Table I, obtained from the containing M bits of redundancy. It has the following form:
simulated single-ended feedforward transconductor, we can
use (5) and (6) to derive the final values we need for the f (x) = 1 for 1 < x ≤ (M + 1)
design, arriving at WSC = 4.135 μm and W = 895 nm. f (x) = 2 · (x − M) for (M + 1) < x ≤ N. (11)
Table I summarizes the state of the fixed components
The first M values encode redundancy into the achieved
parameter values as discussed so far, especially the ideal values
of WFF , WCC , and WSC as derived from (1), (2), and (5). We matching estimate of —the minimum unit of grid-spacing.
can also use this data to confirm that (7) is satisfied. These transconductors are a variation on the idea of an array
of identical minimum sized devices, which are available for
In the next part, we analyze the self-coupled transconductor
width, WSC , and show how to split this component into fixed combination to achieve matching. However, they are a single
and variable components (WSC,fixed ,WSC,variable ). grid-spacing larger than each other, starting from the minimum
sized device. This allows for combinatorial solutions that
can resolve width to within the grid-spacing resolution. The
B. Variable Width Self-Coupled Transconductors remaining N −M devices in the design are sized with multiples
The self-coupled transconductor fixed component, WSC,fixed , of the grid-spacing to achieve a large overall width range with
needs to be a smaller width value than the ideal width, WSC , fewer overall components.
found using (5) so as to allow our variable width self-coupled In general, the choice of the N −1 values for f (x) is design
transconductors to combine with the fixed component to form decisions to be chosen after a process involving tradeoffs
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

NICHOLSON et al.: STATISTICAL DESIGN APPROACH USING FIXED AND VARIABLE WIDTH TRANSCONDUCTORS 5

Fig. 4. Histograms of maximum dc gains obtained across 1024 simulated


OTAs.
Fig. 3. Monte Carlo-based simulations of code yields estimated from
1024 8-bit OTAs in 180-nm CMOS.

for resolution of matching, redundancy, and width range. (negative transconductance not close enough to output
In the 180-nm CMOS process, our prototype was fabricated conductance) or they give a nonzero phase response at dc
in the minimum grid-spacing is  = 5 nm, and the size of (negative transconductance larger in magnitude than output
WSC,variable,1 = 220 nm corresponding to the minimum sized conductance). Given the code-words do not map linearly to
transconductor available. width, the dead-zones are distributed on the code-word axis
in a nonlinear fashion.
Fig. 4 shows the distributions of maximum dc gains sim-
C. Choice of the Number of Bits (N) and Redundancy (M)
ulated for all four design variants. The M = 2 design has
A tradeoff can be found between WSC,fixed and the num- the longest tail of all the distributions, with more nonzero
ber of bits N chosen in the design of the self-coupled bins above 60 dB than any other option. Specifically, it has
transconductors. Fewer bits means a smaller code-space and a 2.94% probability for gains above 60 dB compared with a
a restricted width tuning range requiring a larger WSC,fixed . 3.09% probability for M = 3, and the highest probability for
More bits allows more possible grid-spaced (and overlapping gains above 65 dB at 1.05%. The M = 2 design has average
redundant) width solutions, more width tuning range and a maximum dc gain of 46.9 dB, whilst the M = 3 design has
smaller WSC,fixed at the expense of a larger code-space. After an average maximum dc gain of 47.4 dB.
settling upon values of N, M, and WSC,fixed , we can use (9),
(10), and (11) to calculate the solutions for the variable width
D. Estimation of Width Range of High Gain Solutions
component and the associated high gain code-words.
and Expectation of Grid-Spaced DC Gain
In our presented design, we have chosen to use N = 8,
which corresponds to a code-space size of 256 possible code- Estimating the 99.9% confidence interval (CI) of the mean
words. In practice, these decisions are the result of an iterative of the ratio of output conductance to transconductance,
design approach. During such a design process, and assuming μ(gds/gm ), allows us to derive the range of possible width
a relatively small code-space, we can conduct Monte Carlo differences that correspond to solutions for the maximum
mismatch and process variation statistical trials over the entire dc gain for the digital Nauta OTA. We can derive the CI
code-space, iterating over the values of WSC,fixed and M to see from the ratio of the mean of two random variables using
the effects of varying the fixed component and redundancy Fieller’s theorem, from the mean and variance of the two sets
upon total parametric yield and statistical metrics of average of simulated data (output conductance and transconductance
number of yielding code-words for a given dc gain specifica- curves). Note that this analysis for uncertainty in the ideal
tion and the average maximum dc gain. width is only considering variations in a single feedforward
We simulate four different OTA designs using N = 8, transconductor relative to the entire circuit.
1 ≤ M ≤ 4, and WSC,fixed = 3 μm, across the entire We can then derive from (6) two sets of equations to
code-space using Monte Carlo mismatch and process trials, determine the maximum and minimum ideal width difference
establishing 1024 virtual OTAs. We define code-word yield as from the maximum and minimum values of ζ, ξ
  
the number of OTAs that this particular code-word achieves the gds
dc gain specifications on. Fig. 3 shows the code-word yields Wmax : ζ −1 = 1 + max μ
g
simulated for each of the four possible design decisions for   m
gds
the variable width self-coupled transconductors. The N = 8, ξ = ζ · max μ
g
M = 2 design has 25 code-words, which achieve a 15% or   m 
gds
higher yield, which is only surpassed by N = 8, M = 3, Wmin : ζ −1 = 1 + min μ
g
which has 29 code-words above 15%.   m
Dead-zones in Fig. 3 are related to the fact that gds
ξ = ζ · min μ . (12)
either these code-words do not produce enough dc gain gm
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS

TABLE II III. U NITY-G AIN F REQUENCY M ODEL


PARAMETER VALUES FOR E STIMATION OF R ANGE OF H IGH G AIN
S OLUTIONS AND E XPECTATION OF G RID -S PACED DC G AIN Following the treatment in [8], we provide a model of the
unity gain frequency from the analysis of the capacitance of
the transconductors. Unlike with a digital Nauta OTA made
completely from programmable DTCs from the previously
cited 60-bit 65-nm OTA, the fixed width transconductors used
here provide a simpler analysis. The fixed width transconduc-
tor capacitance model can be written as follows, for nMOS
transistors only, without the disabled state given that the
TABLE III entire OTA is powered OFF when the P signal is low. In the
VARIABLE C OMPONENTS S UMMARY FOR P RESENTED OTA D ESIGN following, we write capacitances Cgd , Cdb , Cgs in terms of
unit transistor widths, that is, normalized to total FF device
width, WFF . We use the definition of k from (2)
WCC
Ctotal,nMOS,fixed = (Cgd + Cdb ) + (2 · Cgd + Cgs ) ·
WFF
WSC,fixed
+ (Cgs + Cdb ) ·
W
 FF 
WSC,fixed
Ctotal,nMOS,fixed = (1 + 2 · k) · Cgd + k + · Cgs
WFF
 
WSC,fixed
+ 1+ · Cdb . (14)
WFF
This range of solutions for the width difference means that
there is a corresponding range of optimal WSC,variable . Given Using our nMOS:pMOS ratio of 1:4 as previously given, our
we have a known range for the ideal width difference, we total capacitance for fixed width transconductors is as follows:
can know write an expression for dc gain in terms of the
Ctotal,fixed = 5 · Ctotal,nMOS,fixed
minimum unit of transconductance uncertainty relative to the  
feedforward transconductors. That is, we estimate the tunable WSC,fixed
Ctotal,fixed = 5 · (1 + 2 · k) · Cgd + 5 · k + · Cgs
transconductance error in our grid-spaced system as the differ- WFF
 
ence between the estimated largest and smallest optimal width WSC,fixed
+5· 1+ · Cdb . (15)
difference between self- and cross-coupled transconductors. WFF
Following on from (1), we find that:
The equation for unity gain frequency follows:
2 · WFF
E[ Av ] = . (13)
Wmax − Wmin f T = gm,unit
   
Inspecting the width range for the variable component of WSC,fixed
· 2π · 5 · (1+ 2 · k) · Cgd + 5 · k + · Cgs
the N = 8, M = 2 design we find that the coverage of WFF
widths that is produced by sizing the transconductors using   −1
WSC,fixed
(10) and (11) provides eight code-word solutions mapping +5· 1+ · Cdb . (16)
WFF
continuously between 1.125 and 1.145 μm in  nm steps.
WSC,fixed =3 μm combines with a value within this range to We combine our values from Table I, estimates for the
form an exact solution to (9). smallest-width transistor capacitances on our 180-nm CMOS
We show the values of the maximum and minimum estimate process at Cgd = 155 aF, Cdb = 195 aF, and Cgs =
for μ(gds/gm ) for a 99.9% CI, the range of W this corre- 222 aF, assuming Cox = 8.4 fF/μm2 ,C j = 0.96 fF/μm2 ,
sponds to, and the expected high gain using the grid-spacing Cjsw = 0.27 fF/μm and unit transconductance estimated at
resolution matching from (13) in Table II. gm,unit = 136 μS. Using these values (16) gives 4.7 GHz for
Our design calculations are completed with this analysis— the maximum unity gain frequency. This result does not take
concluding that the N = 8, M = 2 option is a satisfactory into account the grid-spaced DTCs, as these are much smaller
tradeoff in terms of redundancy and width range of the four than the fixed components, and thus will overestimate the
options—it has a continuous mapping of widths around the maximum unity gain frequency. The ideal matched simulation
estimated range of possible solutions, is expected to achieve of the entire OTA’s unity gain frequency gives a figure of
high average maximum dc gains and has a large number of 4.6 GHz—refer to Fig. 5 showing the simulated gain and
high-yielding code-word solutions. Table III summarizes the phase response at code-word 202—a representative code-word
variable component design decisions thus far, and gives the of high dc gain and stable phase response. The gain curve is
high gain code-words (in decimal format) as found by (9) at the top curve, the phase curve is the bottom curve, showing a
the ideal width (no mismatch effects) and the width range simulated −3 dB gain of 80 dB at 304 kHz and a unity gain
found via (12). frequency of 4.6 GHz.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

NICHOLSON et al.: STATISTICAL DESIGN APPROACH USING FIXED AND VARIABLE WIDTH TRANSCONDUCTORS 7

Fig. 5. Open-loop dc gain and phase response for 8-bit 180-nm digital
OTA—simulated at SC code-word 202.

TABLE IV
S IMULATED C ODE -W ORD Y IELDS W ITHIN THE P REDICTED
I DEAL VARIABLE W IDTH R ANGE

Fig. 6. Distribution of dc gains from fixed width Nauta OTA simulated from
1024 Monte Carlo mismatch and process variation trials in 180-nm CMOS.

We find that each yielding OTA has on average ten-code


solutions, with an average minimum dc gain of 37 dB, with a
standard deviation of 1 dB, and an average maximum dc gain
of 47 dB, with standard deviation of 5 dB, as sampled over the
code-word solution set for all yielding OTAs. The maximum
dc gain achieved was 74 dB. This average maximum dc gain
IV. S IMULATED C ODE -S PACE S TATISTICS of 47 dB is an 11-dB increase over the prediction in (1),
OF N = 8, M = 2 D IGITAL OTA whilst the average minimum is the just above the value of the
We analyze the digital Nauta OTA as presented in Fig. 2 initial gain specification of 36 dB, which was used to size the
using statistical techniques – Monte Carlo process variation feedforward transconductors. The width range of maximum
and mismatch simulations – from which we can quantify the gain code-words corresponds to an implied WSC,variable in the
performance across code-space over a large number of sim- range from 620 to 1300 nm. This range covers the ideal value
ulated OTAs. This allows us to make statistically significant of WSC,variable found in the Table III of 1135 nm.
statements about total parametric yield, individual code- Finally, by the way of comparison, we note that a nonpro-
word yield, and average number of code-word solutions per grammable fixed width version of the OTA using the values
OTA—as estimated across the entire code-space. Furthermore, of WFF , WCC , WSC as listed in Table I only achieves a mean
we can contrast our statistically estimated best code-word dc gain of 44 dB with standard deviation of 10 dB, across
solutions with our high gain codes found via mathematical 1024 Monte Carlo mismatch and process variation trials, with
analysis in Section III, and give average minimum and a yield of 79% for the given gain specification of 36 dB.
maximum dc gains achievable in the code solution set This yield result assumes that no active biasing solutions are
according to the specifications found across all OTAs. used—which, however, can be attempted using a separate
We conducted 1024 Monte Carlo trials for every code-word tunable power supply for certain transconductors to tune the
in the code-space (256), for a total of 262 144 mismatch and transconductance and maximize gain. Fig. 6 shows the distrib-
process variation simulations. Due to our control over the mis- utional shape of dc gains in this fixed width nonprogrammable
match run sequence, we can ensure that we can equivalently Nauta OTA.
treat this as simulating every code-word over 1024 different
OTAs. First, total yield is calculated from the number of OTAs V. C ALIBRATION OF THE P OSITIVE -F EEDBACK
that had no code-words that satisfied the dc gain specification. G AIN -E NHANCEMENT OTA
With 15 from 1024 OTAs with no code-words achieving the Our approach to calibrating the digital Nauta OTA is
gain specification, we estimate total yield at 98.5%. Second, presented previously in [8] and [19]. This digital approach
individual code-word yield can be calculated from the number requires on-chip calibration devices, which can locate the
of OTAs that this particular code-word achieves the gain speci- maximum dc gain through a test voltage sweep, which detects
fications on. Table IV shows the code-word yields correspond- for the presence and width of hysteresis. The presence of
ing to the predicted ideal self-coupled width range identified hysteresis is co-located with the point of maximum dc gain,
in Table III. The ideal solution (code-word 99) calculated and the digital procedure finds the maximum acceptable
analytically under no mismatch conditions is in the middle of dc gain, whilst minimizing for the presence of hysteresis.
this range, with a high yield, and as shown in Fig. 3, second Currently, this requires the OTA to be calibrated in the
top panel is in the middle of a cluster of high yield code-words. foreground periodically.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS

mode, (VDD /2), whilst keeping the differential output voltage


floating and measuring the differential output current. The
slope of the differential output current relative to the input
voltage range is the estimated transconductance.
Output conductance is measured by keeping the differential
input voltage at the common mode, sweeping the differential
output voltage around the common mode, and measuring the
output current. The slope of the differential output current rel-
ative to the output voltage is the estimate output conductance.
Both of these curves are interpolated by a linear regression
giving the estimated slopes and their standard errors.
We report the achieved open-loop dc gain from these two
derivatives and the standard error of the output conductance
using a conservative estimation function, to give the lower
bound of a 95% CI
 
Fig. 7. Layout of the 180-nm digital Nauta OTA occupying 2400 μm2 , not
μgm σgds
Av = 1− (17)
including the digital registers. μgds μgds
where μgm is the measured mean transconductance, μgds is the
In contrast, the calibration approach in measured mean output conductance, and σgds is the standard
Mondal and Krishnapura [20] introduces automatic on-chip deviation of the output conductance as measured by the stan-
conductance tracking using test voltages and a differential dard error reported by the linear regression function. We ignore
difference error amplifier. This introduces another feedback- the standard error of the transconductance as in practice,
path into the positive-feedback gain-enhancement OTA and this is much smaller relative to its mean that the output
architecture that of the addition of a feedback path between conductance. This is due to the fact that the transconductance
the differential output currents and the negative conductance values are not being changed appreciably over code-space,
load. It requires a clocked chopper operation on the new since we have a large fixed width feedforward transconductor
feedback path to alleviate device mismatch effects in the irrespective of code-word, and second, that output conduc-
threshold voltage of all the transconductors. This online tance, as we approach the condition of maximum gain, is
style of calibration is attractive option; however, a clocked approaching a theoretical value of zero. This measurement
averaging circuit to reduce threshold voltage mismatch effects in practice contains more variance than the transconductance,
can only achieve a modest increment in average dc gain, as we approach the noise floor of measurement setup.
since it is a passive mismatch cancelation technique [20]. Transconductance and output conductance measurements
Future work on digital calibration includes strategies for were conducted across the entire code-space of each OTA and
classifying the gain directly, rather than assuming colocation repeated three times. Linear dc gain is calculated by (17), then
with hysteresis, and methods of comparing code-words. averaged across the three experiments for each code.
Further techniques for background calibration will be required Average power consumption when driven by a VDD /2
for continuous output availability and linearization of voltage peak-to-peak differential sine wave in open-loop configuration
to current conversion will be beneficial in the embedding of was 2.4 mA. This type of OTA structure has bias currents
the OTA into larger digitally assisted analog circuits. dependent on the input signal. Thus, we estimate power at
the maximum expected range of the input signal amplitude.
Measured differential transconductance over ten prototype
VI. E XPERIMENTAL R ESULTS OF N = 8, M = 2 chips had a mean of 1.9 mS, standard deviation of 22 μS,
D IGITAL NAUTA OTA P ROTOTYPES at the maximum gain code-word, implying a single-ended
We show the layout of the prototype digital OTA in Fig. 7 transconductance of the feedforward transconductors of
in lieu of a micrograph of the die. This shows the analog 3.8 mS as compared with the design specification of 4 mS.
components only, without the digital registers holding the Maximum standard error of any individual mean
programming code-word. Visible on the left are the three fixed transconductance measurement across the maximum gain
width transconductor components, and on the right is the array code-word was 0.03%.
of eight digitally programmable transconductors. The chips Average maximum dc gain across the ten chips was 60 dB
were supplied with a power supply of VDD = 1.8 V. with the 95% CI calculated to be 54–64 dB. The maximum
Transconductance and output conductance curves were gain codes corresponded to an ideal total variable width
measured on each prototype chip’s OTA (refer to Fig. 2) solution range of 1070–1475 nm, compared with the analytical
via a programmable two-panel source measurement unit solution of 1135 nm. The ten prototype chips had an average
utilizing a six-wire interface—four wires for the differential of 25 code-words per OTA satisfying the gain requirement,
input and ground and two wires for the differential output. compared with our simulated average value of ten code-words.
Transconductance was measured with the differential input Figs. 8 and 9 show a heat-map of dc gains measured
being swept across a range of voltages around the common across all ten prototype chips, for two ranges of code-space.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

NICHOLSON et al.: STATISTICAL DESIGN APPROACH USING FIXED AND VARIABLE WIDTH TRANSCONDUCTORS 9

Fig. 8. Heat-map showing measured dc gains across ten prototype chips, inset 1280 nm to 1590 nm.

These heat-maps have been converted from code-word to in Fig. 11. Fig. 11 shows the long tail distribution of gains as
implied total variable transconductor width and sorted for we see in the simulated digital OTAs represented in Fig. 4 and
nonmonotonic increasing width. In some cases, due to the in the fixed width Nauta OTA simulated by process variation
nature of the encoding, some widths are duplicated, since some and mismatch Monte Carlo trials represented in Fig. 6. This
code-words map to the same total variable transconductor long tail is the behavior we want and expect such that a
width, and in other cases, there are gaps. These heat-maps calibration procedure can trim the amplifier into a high gain
clearly show the two bands of high dc gains around widths code-word.
1475 and 1135 nm. This shows that there are horizontal bands
of high dc gain solutions across the sampled prototype chips, VII. T RADITIONAL OTA C OMPARISON
which indicates consistent code-word solutions across any As a reference point to compare against our variation-aware
particular tape-out of the OTA. OTA architecture, we simulate an ac optimized traditional
Fig. 10 shows the estimated code yield across code-space as analog OTA design in the 180-nm CMOS process as the digital
measured across ten prototype chips. Due to the small sample Nauta OTAs presented. The circuit is given by Binkley and the
size, these code yields may be overestimating the achievable OTA schematic is given in Fig. 12 [11].
code yields. To more accurately statistically quantify code All devices are minimum length as similar to the digital
yield and overall chip yield, we would need to test more Nauta OTAs under consideration, with widths chosen for the
prototype chips to build up confidence though a larger sample fastest traditional OTA circuit. This standard input differential,
size. However, the overall shape of the curve is similar to single-ended output OTA design has a mean of 2.9 GHz unity
the simulated one as shown in Fig. 3 second panel from gain bandwidth, a mean dc gain of 49 dB with standard
the top (M = 2). The two codes that are shown as 100% deviation of 2 dB—as simulated across 1024 mismatch and
yield correspond to code-word 99 (width 1135 nm), which process variation Monte Carlo statistical trials, refer to Fig. 13.
is the ideal code-word predicted from analysis in Section III Fig. 13 shows the yield as measured by a gain specification
(Table IV) as well as code-word 199 (width 1475 nm). Both of 36 dB corresponding to the example in Section II. Note
of these codes we measured as having dc gain greater than the that in comparison with the shape of dc gain distribution for
specification across all ten chips. the Nauta OTA, the traditional distribution is approximately
Finally, we give the histogram of experimentally measured normally distributed around the mean gain—there is no long
open-loop dc gains across all ten prototype digital OTAs tail of high dc gains like in Figs. 4 and 6.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS

Fig. 9. Heat-map showing measured dc gains across ten prototype chips, inset 1030 nm to 1205 nm.

Fig. 10. Code yields as experimentally measured across the ten prototype Fig. 11. Histogram of experimentally measured open loop dc gains across
chips. all ten 8-bit 180 nm digital Nauta OTAs.

Core power consumption is 810 μW using a 1.8 V supply,


including voltage reference circuits, with an external current This contrasts against our digital Nauta OTA, which con-
reference at 50 μA. Silicon area is estimated at 1305 μm2 — sumes 4.5 mW using a 1.8 V supply, occupies 2400 μm2 in
not including the required current reference and voltage total (not including the digital registers), has a 4.6 GHz unity
bias generation circuits. Binkley reported tape-out area at gain bandwidth and an average maximum dc gain of 60 dB.
1900 μm2 . Slew rate is limited by the small maximum output This is a five times higher gain-bandwidth product than the
current of twice the bias current set—in this example 100 μA. traditional OTA design. Expected differential output voltage
Usable output voltage range is approximately 600 mV. This range is approximately 1 V using (2·VDD −4·VT −4·VDS,sat)—
design can only be calibrated for process variations postlayout this is enough voltage headroom to maintain both pMOS and
via adjusting dc bias conditions, and mismatch variations nMOS drivers turned ON whilst keeping the tristate transistors
cannot be corrected. still in strong inversion, using nominal values of VT = 420 mV
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

NICHOLSON et al.: STATISTICAL DESIGN APPROACH USING FIXED AND VARIABLE WIDTH TRANSCONDUCTORS 11

TABLE V
C OMPARISON OF P UBLISHED CMOS OTAs W ITH T HIS PAPER . * N EXT TO
FoM VALUE I NDICATES A S IMULATION -O NLY R ESULT

of the digitally programmable variable width transconductor


Fig. 12. Traditional OTA design in 180 nm CMOS simulated as comparative ensures that we are robust to variations, and gives the calibra-
reference design. From [11] © Wiley 2008. Reproduced with permission. tion circuit enough code-space to find solutions to the device
matching problem.
Our results show that the grid-spacing strategy implies an
order of magnitude matching resolution improvement in max-
imum dc gain compared with the first-order dc gain estimate.
The prototype chip results show an average maximum gain
of 60 dB compared with the first-order dc gain estimate of
36 and 47 dB average maximum as simulated across process
and mismatch Monte Carlo trials.
Table V shows our measured results against previously
published work. The digital Nauta OTA is a high-speed, high
dc gain, and scalable solution for digitally assisted analog
OTA design with a state-of-the-art FoM.

R EFERENCES
[1] B. Murmann, P. Nikaeen, D. J. Connelly, and R. W. Dutton, “Impact of
scaling on analog performance and associated modeling needs,” IEEE
Trans. Electron Devices, vol. 53, no. 9, pp. 2160–2167, Sep. 2006.
Fig. 13. DC gain distribution in Monte Carlo mismatch and process variation [2] T. Ytterdal and C. Wulff, “On the energy efficiency of ana-
trials of a traditional analog OTA design (given in Fig. 12) as simulated in log circuits in nanoscale CMOS technologies,” in Proc. Int. Conf.
180-nm CMOS. Microelectron. (ICM), Dec. 2008, pp. 240–243.
[3] P. G. Drennan and C. C. McAndrew, “Understanding MOSFET mis-
and VDS,sat = 200 mV for 180-nm CMOS. It can be calibrated match for analog design,” IEEE J. Solid-State Circuits, vol. 38, no. 3,
pp. 450–456, Mar. 2003.
for mismatch and process variations postlayout by digital [4] A.-J. Annema, B. Nauta, R. van Langevelde, and H. Tuinhout, “Analog
programming and by adjusting dc bias conditions if required. circuits in ultra-deep-submicron CMOS,” IEEE J. Solid-State Circuits,
Thus, our digital Nauta OTA is capable of faster slew rates, vol. 40, no. 1, pp. 132–143, Jan. 2005.
[5] P. R. Kinget, “Device mismatch and tradeoffs in the design of analog
higher output voltage ranges, higher dc gains, and higher unity circuits,” IEEE J. Solid-State Circuits, vol. 40, no. 6, pp. 1212–1224,
gains at similar silicon areas with these advantages improving Jun. 2005.
in finer line-widths as the CMOS process scales down. [6] B. Murmann, “Digitally assisted analog circuits,” IEEE Micro, vol. 26,
no. 2, pp. 38–47, Mar. 2006.
[7] A. N. Irfansyah, A. P. Nicholson, A. Iberzanov, J. Jenkins, T. Lehmann,
VIII. C ONCLUSION and T. J. Hamilton, “Automatic tuning of digitally-controllable positive-
feedback OTAs in continuous-time sigma–delta modulators,” Analog
The positive-feedback gain-enhancement OTA produced Integr. Circuits Signal Process., vol. 89, no. 2, pp. 469–483, Nov. 2016.
using a statistical design framework to enable variation-aware [8] A. P. Nicholson, A. Iberzanov, J. Jenkins, T. J. Hamilton, and
T. Lehmann, “A statistical design approach for a digitally programmable
digital programmability is shown to be a viable solution mismatch-tolerant high-speed Nauta structure differential OTA in 65-nm
for digitally assisted analog circuit design scaling into deep CMOS,” IEEE Trans. Very Large Scale Integr. (VLSI) Syst., vol. 24,
submicron CMOS. The digital Nauta OTA presented here is no. 9, pp. 2899–2910, Sep. 2016.
a high-speed single-stage OTA capable of high dc gains, high [9] A. Nicholson et al., “A 0.3mm2 10-b 100MS/s pipelined ADC using
Nauta structure op-amps in 180nm CMOS,” in Proc. IEEE Int. Symp.
unity gain frequencies, and well placed for scaling into smaller Circuits Syst. (ISCAS), May 2013, pp. 1833–1836.
process nodes. [10] A. N. Irfansyah, L. Pham, A. Nicholson, T. Lehmann, J. Jenkins, and
Our technique of using a tradeoff between redundancy and T. J. Hamilton, “Nauta OTA in a second-order continuous-time delta-
sigma modulator,” in Proc. IEEE 57th Int. Midwest Symp. Circuits
range, through grid-spacing width differences in the design Syst. (MWSCAS), Aug. 2014, pp. 849–852.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

12 IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS

[11] D. Binkley, Tradeoffs and Optimization in Analog CMOS Design. Astria Nur Irfansyah received the B.Eng. degree in
Hoboken, NJ, USA: Wiley, 2008. electrical engineering from Gadjah Mada University,
[12] T. McConaghy, K. Breen, J. Dyck, and A. Gupta, Variation-Aware Yogyakarta, Indonesia, in 2004, the M.Eng. degree
Design of Custom Integrated Circuits: A Hands-on Field Guide. in electrical engineering from the University of
New York, NY, USA: Springer, 2012. New South Wales, Sydney, NSW, Australia, where
[13] J. Maunu, M. Pankaala, J. Marku, J. Poikonen, M. Laiho, and A. Paasio, he is currently pursuing the Ph.D. degree with the
“Current source calibration by combination selection of minimum sized School of Electrical Engineering and Telecommuni-
devices,” in Proc. IEEE Int. Symp. Circuits Syst. (ISCAS), May 2006, cations, with a research focus on digitally-assisted
pp. 549–552. sigma delta modulation circuits.
[14] K. Ragab, R. Gharpurey, and M. Orshansky, “Embracing local variability He has been on leave from the Institut Teknologi
to enable a robust high-gain positive-feedback amplifier: Design method- Sepuluh Nopember, Surabaya, Indonesia, where he
ology and implementation,” in Proc. 13th Int. Symp. Quality Electron. is currently an Academic Staff with the Department of Electrical Engineering.
Design (ISQED), Mar. 2012, pp. 143–150.
[15] B. Nauta, Analog CMOS Filters for Very High Frequencies (The Kluwer
international series in engineering and computer science: Analog circuits Julian Jenkins received the B.Sc. degree in
and signal processing). Norwell, MA, USA: Kluwer, 1993. computer science and the B.Eng. degree in
[16] A. N. Irfansyah, A. Nicholson, J. Jenkins, T. J. Hamilton, and electrical engineering from the University of
T. Lehmann, “Subthreshold operation of Nauta’s operational transcon- New South Wales, Sydney, NSW, Australia, in
ductance amplifier,” in Proc. IEEE 13th Int. New Circuits Syst. 1996 and 1997, respectively, where he is currently
Conf. (NEWCAS), Jun. 2015, pp. 1–4. pursuing the Ph.D. degree.
[17] A. Nicholson, J. Jenkins, A. van Schaik, T. J. Hamilton, and T. Lehmann, From 1997 to 2000, he was with Quality
“A digital to transconductance converter for Nauta structure op-amps Semiconductor Australia, Sydney, where he is
in 65nm CMOS,” in Proc. IEEE 57th Int. Midwest Symp. Circuits involved in working on 100-Mbps Ethernet.
Syst. (MWSCAS), Aug. 2014, pp. 173–176. In 2000, he joined Hiband Semiconductor, later
[18] B. Nauta, “A CMOS transconductance-C filter technique for very high acquired by Cypress Semiconductor to focus on
frequencies,” IEEE J. Solid-State Circuits, vol. 27, no. 2, pp. 142–153, 3.5 Gbps SerDes. In 2005, he co-founded Perceptia Devices with a focus
Feb. 1992. on SerDes and PLLs in the 10-28-Gbps range. His current research interests
[19] A. Iberzanov, A. Nicholson, J. Jenkins, T. Lehmann, and T. J. Hamilton, include designing analog circuits in sub 65-nm technologies with a focus on
“Calibration of the Nauta structure differential OTA,” in Proc. IEEE Asia SerDes, PLLs, DCOs, CDRs, and filters.
Pacific Conf. Circuits Syst. (APCCAS), Nov. 2014, pp. 189–192.
[20] I. Mondal and N. Krishnapura, “Gain enhanced high frequency OTA
with on-chip tuned negative conductance load,” in Proc. IEEE Int. Symp.
Circuits Syst. (ISCAS), May 2015, pp. 2085–2088. Tara Julia Hamilton (S’97–M’00) received the
[21] S. Chen, L. He, and L. Zhang, “High gain, high speed OTA for S/H B.E. degree (Hons.) in electrical engineering and
circuit in 14-b 100-MS/s pipeline ADC,” in Proc. 13th Int. Symp. Integr. the B.Com. degree from the University of Sydney,
Circuits (ISIC), Dec. 2011, pp. 254–257. Sydney, NSW, Australia, in 2001, the M.Sc. degree
[22] W. Yan, R. Kolm, and H. Zimmermann, “A low-voltage low-power fully in biomedical engineering from the University of
differential rail-to-rail input/output opamp in 65-nm CMOS,” in Proc. New South Wales, Sydney, in 2003, and the Ph.D.
IEEE Int. Symp. Circuits Syst. (ISCAS), May 2008, pp. 2274–2277. degree from the University of Sydney in 2008.
[23] H. Uhrmann, F. Schlogl, K. Schweiger, and H. Zimmermann, She is currently a Senior Research Lecturer in
“A 1GHz-GBW operational amplifier for DVB-H receivers in 65nm biomedical engineering and neuroscience, MARCS
CMOS,” in Proc. 12th Int. Symp. Design Diagnostics Electron. Circuits Institute, Western Sydney University, Penrith, NSW,
Syst. (DDECS), Apr. 2009, pp. 182–185. Australia. Her current research interests include neu-
[24] O. Abdelfattah, G. W. Roberts, I. Shih, and Y.-C. Shih, “An ultra- romorphic engineering, mixed-signal integrated circuit design, and biomedical
low-voltage CMOS process-insensitive self-biased ota with rail-to-rail engineering.
input range,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 62, no. 10,
pp. 2380–2390, Oct. 2015.
Torsten Lehmann (M’94–SM’06) received the
M.Sc. and Ph.D. degrees in electrical engineering
from the Technical University of Denmark, Lyngby,
Denmark, in 1991 and 1995, respectively, with a
focus on selflearning VLSI neural networks.
He spent two years as a Research Fellow with
the University of Edinburgh, Edinburgh, Scotland,
where he was with biologically inspired artificial
neural systems. From 1997 to 2000, he was an
Assistant Professor in electronics with the Technical
Andrew Peter Nicholson received the B.Eng. University of Denmark, where he was involved in
degree (Hons.) in computer engineering from the low-power, low-noise, low-voltage analog, and mixed analog-digital integrated
University of Newcastle, Callaghan, NSW, Australia, circuits. From 2001 to 2003, he was a Principal Engineer with Cochlear Ltd.,
the M.Eng. degree in electrical engineering from Sydney, NSW, Australia, where he was involved in the design of the world’s
the University of Sydney, Sydney, with specializing first fully implantable cochlear implant. He is currently an Associate Professor
in wireless. He is currently pursuing the Ph.D. in microelectronics with the University of New South Wales, Sydney, NSW,
degree with the School of Electrical Engineering Australia. He has authored over 100 journal papers, conference papers, book
and Telecommunications, at the University of chapters, and patents in integrated circuit design for a range of applications.
New South Wales, Australia, with a focus on His current research interests are in solid-state circuits and systems (analog
researching alternative analog design strategies in and digital), biomedical microelectronics, ultralow temperature electronics,
nanometer CMOS. nanometer CMOS, and green electronics.

You might also like