A Calibration concept for SAR ADC for operation in LAr...SAR ADC Architecture 9/11/2018 Yuan Mei 3 DAC SAR Logic Vin Digital Outputs Comparator •Bottom plate sampling method with

Name: Yuan Mei

Brookhaven National Laboratory

Instrumentation Division

ULITIMA 2018

A Calibration concept for SAR ADC for operation in LAr TPC

9/11/2018 Yuan Mei 1

[email protected]

Motivation

▪Design a 12-bit ADC running 2MS/s for DUNE

▪Risk mitigation for cold temperature @77K

▪Errors from analog circuits should be overcome with digital calibration, transferring complexity from analog to digital domain

9/11/2018 Yuan Mei 2

SAR ADC Architecture

9/11/2018 Yuan Mei 3

DAC SAR Logic

Vin

Digital Outputs

Comparator

• Bottom plate sampling method with synchronous timing

• 12-bit resolution with 14 conversion-cycle (2-bit redundancy)

• Digital foreground calibration implemented

Static Error sources

• Conversion accuracy is subject to the non-idealities of analog components, the main error source are DAC mismatch and comparator offset errors.

• SAR only has one comparator, offset won’t affect linearity.

• Bottom-plate sampling is performed, thus injected charge to the top-plate is independent of input signal and contribute a fixed offset

• Auto-zeroing and chopper techniques are often used to eliminate comparator offset

• Capacitor mismatch is fatal to ADC performance if not solved

9/11/2018 Yuan Mei 4

Capacitor mismatch issues

• 12-bit matching is needed in the capacitor array

oRandom mismatch: Gaussian distribution

oGradient mismatch: Location-dependent

oRouting mismatch: Layout dependent

• Unit-cap and common-centroid layout can fix gradient mismatch and routing mismatch

• Random mismatch can be solved by using larger unit-capacitor [1] or using calibration method

9/11/2018 Yuan Mei 5

Capacitor mismatch in DAC

9/11/2018 Yuan Mei 6

DAC mismatch limit ADC performance and calibration would improve ADC!

Idea of Digital Calibration

• Post-processing of codes

oRelies on digital signal processing

oNo feedback to ADC and thus no stability issue

oDoesn’t required complex analog circuit

oDigital circuits scale in advanced process

Correlation-based calibration is selected for this prototype

9/11/2018 Yuan Mei 7

Correlation-based Calibration

• Inject a small amount of analog signal that is uncorrelated with the input signal. Since it passes through the same path as DAC signal, it encounter same non-idealities and hence can detect the mismatch error

• Using a statistical correlator (histogram method) or the least mean square (LMS) algorithm, the inject signal can be “correlated out” in the digital back-end and capacitor mismatch error are estimated in the process [2]

9/11/2018 Yuan Mei 8

Implementation of Algorithm

9/11/2018 Yuan Mei 9

• Each sampled analog signal will be digitized twice

• Decimal output codes d1 and d2 will create a error function

• Error function will provide information to infer the unknown weighting vector W

• Adaptive learning algorithm (LMS) to update weight coefficient in DAC in which error is gradually forced to zero

• Learning procedure converges, the mean of d1 and d2 will yield the correct digital output codes

Fig. Block diagram of the perturbation based digital calibration for SAR ADC [3]

SARCORE

Encode 1

Encode 2

e[n]

LMS

d

a a

2

D1

D2 d2

d1

doutVin

W1

W2

Calibration engine

LMS : Wk [N+1] = Wk[n] - µ*e[n]*dk[n]

e [N] = d1[n] - d2[n] -2 d

Capacitor ratio with redundancy

9/11/2018 Yuan Mei 10

Calibration circuit

9/11/2018 Yuan Mei 11

Advantage• Calibration circuit can be utilized in DAC• Calibration circuit is stable at cold • Calibration control is simple

Disadvantage• Decrease input dynamic range

Notes :Single-ended for simplicityImplemented in differential

Vinput

Bootstrap_Switch

Transmission-Gate Switch

Vref_pos

Transmission-Gate Switch

Vref_neg

1920 1024 544 288 144 80 40 24 12 8 6 2 2 1 10

MSB LSB Calibration Cap

VCM

Controlled by SAR Logic

VCM

Calibration Circuit

Timing Diagram

9/11/2018 Yuan Mei 12

40 Cycles

Reset

Sample

Output1

D14 D13 D12 D2 D1

Clk

Output2

D14b D2b D1b

Input regeneration

• One complete conversion consist of a 5 clock-cycle sampling phase and two 14 clock-cycle conversion phases;• Besides, 1 clock-cycle is reserve for reset and 6 clock-cycle is for input regeneration.

Timing diagram of the proposed calibration method

Simulations with digital calibration

• Simulations are done in Cadence @ 77K temperature

• All blocks are implemented and simulated in transistor level

• Unit capacitor is MIM cap provided by foundry (about 10 fF)

• DAC mismatch errors are the major error source

9/11/2018 Yuan Mei 13

Capacitor mismatch calibration verification

9/11/2018 Yuan Mei 14

• Set random mismatch on capacitor weights

• Simulate to see if calibration can find the optimal weight and improve the performance

• Other mismatch is excluded from this verfication

Capacitor mismatch calibration

9/11/2018 Yuan Mei 15

Simulated with 2^14 (16,384) samples, weights are converged around 10,000 samples X-axis: # of samples Y-axis: Capacitor weight

Capacitor mismatch calibration

9/11/2018 Yuan Mei 16

Simulated with 2^14 (16,384) samples, weights are converged around 10,000 samples X-axis: # of samples Y-axis: Capacitor weight

Capacitor mismatch calibration

9/11/2018 Yuan Mei 17

With calibration, all the weights in the DAC are updated X-axis: # of samples Y-axis: Capacitor weight

Dynamic test

9/11/2018 Yuan Mei 18

FFT points:2^14

Without CalibrationENOB = 8.3 bits

With CalibrationENOB= 11.2 bits

• ENOB is improved• Dynamic range also improved

Dynamic test

9/11/2018 Yuan Mei 19

FFT points:2^17

Without CalibrationENOB = 8.3 bits

With CalibrationENOB= 11.9 bits

• ENOB is further improved from 11.2bits to 11.9bits • Dynamic range also improved from 69.1 dB to 73.6dB

With more samples :

Static test

9/11/2018 Yuan Mei 20

Without Calibration Without Calibration

• Linearity is improved with the digital calibration

Sample points:2^14

A lot missing codes! NO missing code!

Static test

9/11/2018 Yuan Mei 21

With samples 131,072 (2^17)

• Linearity is further improved with more samples

With samples 16,384 (2^14)

Floorplan

9/11/2018 Yuan Mei 22

N-DAC

P-DAC

DummyDummy Dummy

Dummy Dummy Dummy

Du

mm

yD

um

my

Du

mm

yD

um

my

Dummy Dummy

Dummy Dummy Dummy

N-DAC DC Switches

P-DAC DC Switches

VCMSwitch

VCMSwitch

Vin+ BootstrapSwitch

Vin- BootstrapSwitch

Comparator

Preamp

Preamp

NomalModeSAR

Logic

CalibrationModeSAR

Logic

Clock GeneratorData Formatter

+LVDS I/O

Digital DomainAnalog Domain

Floorplan of SAR ADC prototype

Die Photo

9/11/2018 Yuan Mei 23

Die photo of SAR ADC prototype

Measurement

• Chip came back in Mid-July and measurement will be done in October

9/11/2018 Yuan Mei 24

Power consumption @77K(Simulation by PEX)

9/11/2018 Yuan Mei 25

Block Name Power consumption 2MS/s

Digital 240 µw

Analog 280 µW

Total 520 µW

Acknowledgement

• My colleagues : Gabriella Carini, Huchen Chen, Mietek Dabrowski, Shaorui Li , and Emerson Vernon

• Fermi lab collaborators and their cold model for TSMC 65nm at 77K

• DUNE and DoE for support of this work

9/11/2018 Yuan Mei 26

Reference

[1] Yi-long Yu, ADI talk, A 12 bit 100MS/s two step hybrid ADC in 40nm CMOS with statistical calibration

[2] Ahmed M. A. Ali. 2016. High Speed Data Converters. Institution of Engineering and Technology.

[3] W. Liu, P. Huang and Y. Chiu, "A 12-bit, 45-MS/s, 3-mW Redundant Successive-Approximation-Register Analog-to-Digital Converter With Digital Calibration," in IEEE J. of Solid-State Circuits, vol. 46, no. 11, pp. 2661-2672, Nov. 2011.

9/11/2018 Yuan Mei 27

Q & A

Thank you for your attention.

9/11/2018 Yuan Mei 28

Backup slides

9/11/2018 Yuan Mei 29

Wi represents the respective bit weight. µ i is the step size of the update equation and is scaled according to the bit.e[N] is the total error of the Nth step. dN are the raw ADC output;Δd are digitized offset of the inserting analog offset Δa

LMS : Wk [N+1] = Wk[n] - µ*e[n]*dk[n]

e [N] = d1[n] - d2[n] -2 d

Convergence of Error

9/11/2018 30