Ananthprofilepln

Ananth Chellappa

Analog Design Engineer

Professional Profile

Contents

Summary Work experience Examples of Initiative and Innovation Project details Skills Patent Applications Education

Summary

Graduate of IIT(M) and Georgia Tech with 10 years industry experience at Maxim Integrated Products, Texas Instruments and Analog Devices

Analog designer with exposure to PLL, ADC, DAC and switching-regulator design

Good with tools, scripting (perl/unix) and documentation; recognized by peers for speed, thoroughness and proactivity

Fast learner, time-sensitive person, motivated to contribute

Work Experience Jul 2009 – present : Maxim Integrated Products (Phoenix, AZ)

Oct 2005-Apr 2009 : Texas Instruments (Dallas)

- bandgap references, amplifiers, linear-regulators, switching-regulators, oscillator

Jan 2005 – July 2005 : Analog Devices (Tucson, AZ)

- autozero opamp, current reference, low-power clock-multiplier PLL

Aug 2001 – Dec 2004 : Graduate Research Assistant at Geogia Tech

- high-speed optoelectronic receiver (TIA + limiting amp)

July 1998 – Mar 2001 : Texas Instruments India

- PLL, pipeline ADC, DAC/SCF, r-string DAC, SAR ADC

Initiative and Innovation

Issue Action

Custom digital cells (weak, but good for area) library lacks the 2x4 decoder used in the 4x16 decoder of the DAC in the buck-regulator

reference section

Drove standard-cell library team to provide this cell in a new release; impacts all of Maxim

Clock generator used to generate 6 different clock phases uses 22 D-

flip-flops – huge area

Redesigned to use only 6 D-flip-flops and moved to custom cells to save more than 50% area

Reference bias-current takes about 60 us (nom) to reach final value

Redesigned (moved from single-stage amp with large comp-cap to 2-stg amp with small comp cap amd miller-comp) to save area and reduce time-to-final to about 2.5 us worst-case and achieve better

PSRR without VCC filter (25 pF)


Issue Action

To talk to the chip (on the bench) through the proprietary 3-pin test-

interface, designers are using a board with push-buttons to send the required pulses. While this

accomplishes the result of moving from register to register and

sending required data, it does not provide visual feedback and the

ability to automate (through macros). The designer must

remember or write down what data has been sent and what state he is

in

Designed a new board with simple level-shifters (own expense) and interfaced the chip to the PC through an Arduino MCU board and wrote the GUI on the PC with Autohotkey (open-source

Windows scripting utility) – everything done from scratch. This enabled fast bench-correlation and macros to be written to do a force-trim on about 200 samples of a new chip (by the Test-engineer

using the designer’s setup) to sample the customer very early. Also used on a class-D audio-power-

supply boost-regulator project. Both the MAX8989 have been in production successfully

This is an example of doing what needs to be done to achieve success. In the context of my goals related to IC design, doing such interfaces is not a specific interest of mine


Issue Action

When running LPE (parasitic back-annotated) simulations, we find movement on sensitive nodes.

Designers look manually in the cap file for coupling caps to aggressors, but, sometimes, this coupling can

be secondary coupling through other “quiet” nodes

Wrote an ocean script that would look at all available waveforms in the simulation result and

arrange nodes based on amount of movement during the problem window. Helped us save time in locating aggressors and address layout issues

There are thousands of nodes and we’d like to save all of them for debug purposes. However, if we

save waveforms with compression, then we lose accuracy – which is important for the critical signals

Moved to spectre version 12 so that all waveforms could be saved but only selected waveforms could

be saved with compression to yield manageable plotfile sizes


Issue Action

MAX77387 – the flash-LED current-regulator signals the digital the boost output

voltage is not high enough and the digital commands the boost output higher – but, the output voltage does not match the no-load

voltage in the load-line architecture. So, the digital might decide that the boost-output is aleady “too-high” and not increase it further

In this application, the load on the boost is well defined – it is the DAC setting of the Flash-LED current-regulator. This

information can be used to add the required adjustment at the output of the

error-amp of the boost and enhance load-regulation without changing the AC loop-gain and worrying about stability. This is

in the production version

To check for floating nodes, designers run the checker and then manually go through

report file and manually locate each of hundreds of nodes in the schematic

Did the first version and then handed off to EDA a tool that now lets designers

specify the report file and then just hit a “next” button in Cadence to be taken to

each node in the schematic automatically


Issue Action

The simple R-string DAC with LSB taps on top has a transient non-monotonicity issue –

when you traverse MSB codes, you immediately go from the min to the max

switch on the LSB RDAC – on top, but the voltage propagation from the MSB DAC on the bottom takes time – resulting in a glitch

on the output

Add feedforward caps from a few tap points in the MSB DAC to considerably alleviate the problem to the point that the

buck-regulator cannot respond to the glitch because it is so narrow

In meetings, designers tend to talk about circuit details based on their familiarity with

the result that others cannot follow. Capturing the ideas with Visio of Cadence

takes time

Use a sharpie marker to produce good enough, and clear-enough plots that can be scanned and put up on the projector. Not an original idea, but my adoption of

it has since seen my peers follow suit


Issue Action

Was new to Maxim and my first project was on Viewlogic – which is

a very inferior design platform compared to Cadence. A translator

existed to migrate the data to Cadence, but no one had used it

Learned the use of the translator, migrated data to Cadence and pushed management for layout

resources to run LVS to fix translation errors in the translated schematics. This allowed this design to be respun on two separate projects

which have since resulted in significant revenue for the group

Running testmode simulations is a task no one looks forward to and usually, on 2nd and 3rd passes, we

find bugs in testmodes (on silicon) because they weren’t simulated

Changed workflow to generate a push-button ocean script for every simulation and then created

a master perl-script that could stich together an arbitrary number of simulations, run them and also post-process simulation results either by using ocean to look at plots or perl to look at

simulation log output and then report pass/fail


Issue Action

There was no good corner simulation setup and designers were

running the corners they assumed would be the worst-case

Learnt the use of the new Maxim in-house tool and held two training demos to train my peers;

This tool is now widely used at our design center

Sometimes it is necessary to access a signal buried within the hierarchy from the testbench level. In the past, we have had to propagate this node to the top-level by putting pins on

each cell in the hierarchy

Found deepprobe on the Cadence support site, learnt its use and fanned out to the team. It is currently widely used for different purposes – triggering faults, injecting offsets, speeding up

startup in simulations, etc


Issue Action

Customer (Infineon) had supply volage to the PA step-down-

regulator (MAX8989) as high as 5.5V. In this condition, though they

preferred to have the buck in dropout to avoid switching noise, their DAC (driving the reference

(REFIN) input could not produce a high enough voltage to make this

happen. So they failed the RF spur test

Proposed a new feature wherein a REFIN higher than 1.8V would send the buck and bypass-LDO

into very high-gain mode (effectively guaranteeing dropout) to prevent switching.

Normal operation where the buck-output was being modulated by REFIN for efficiency reasons

did not require REFIN above 1.5VThis feature was in the final production version

Block Design SummaryBlock Schematic only Working silicon Production Institution(s) Remarks

PLL 1 2 TI-I, ADI Maneatis style

Bandgap 2 1 TI-Dallas 2 precision designs

Autozero amp 1 1 ADI, TI-Dallas

Sw-cap gain stage 2 1 TI-I, TI-Dallas 1 involved amp

design

Performance linear amp 1 TI-Dallas TIA for TS AFE

Linear regulator 2 2 1 TI-Dallas

Switching regulator

1 1 TI-Dallas Reuse, no core changes

2 MaximDetailed design of boost; reuse of hyst-buck

Voltage buffer 2 2 TI-Dallas

ADC 1 TI-I Recyclic

DAC 1 TI-I R-string

Project Detail – Dual Phase Boost Sept. 2010 to Nov. 2011 with diversions from re-spins of

MAX8989 PA buck and MAX77693 – Maxim’s PMIC in the Galaxy III

Target application – camera flash-LED driver. Motivation : 2 inductors carrying half the current each carry ¼ the energy of a single inductor carrying all of the current – results in significantly lower footprint – 70% smaller than competing 1.5A solution

Design activity influenced all blocks; Lowest minimum-ON-time boost designed within the group and only one with a true PWM mode; world’s smallest 2A boost solution

Project Detail – Touchscreen AFE

July 2008 – Jan 2009. 90 nm CMOS. ASIC device with digital owned by customer Joined team July 2008 to work on new device. Silicon for

current device in eval/debug Customer – large consumer electronics co. Initially worked on internal LDO for system oscillator. Did

preliminary design (from scratch). Later ported (in-progress) a 65 nm design from another group

Soon began work on RX architecture, filter topology and transimpedance amp (TIA) frontend


• Worked on internal (on-chip cap) LDO for system oscillator

• Designed transimpedance amp frontend for new device

• Drove RCA of LDO high-temp failure

• RCA of LDO transient issue; LDO fixes for new device

• Explained yield loss in RX-SNR test @ 100 kHz for current device

• Drove RCA of 0.2% of units powerup fail in system

• Identified misc improvements to help area/power

• Misc tasks to support project


• New TIA design supported 4x original frequency

• Lower noise at same power obtained by increasing VDSATs of active load devices and topological modification of first stage

• VREF is divided down version of the ANA LDO voltage. Due to nature of the LDO design, noise depended on mismatch, and flicker noise was significant at 100 kHz – explained the loss of yield in the SNR test at 100 kHz

Project Detail – Notebook Charger

• 2Q 2008; Joined project which was significantly behind schedule

• Took over system oscillator from another resource; found that resistor area in the circuit could be reduced substantially

• Helped co-worker design a low-offset amp

• Handled entire verification of supervisory blocks and high-voltage level-shifter

• Key role in top-level simulation efforts

Project Detail – GPS PMIC

• 2Q 2007 – 1Q 2008

• Aggressive schedule; reuse of catalog blocks; chip: pwr mux, battery charger, resistive touchscreen controller, buck regulators, white LED boost regulator, LDOs

• Complete ownership of buck-regulators (handled import, verification, minor tweaks, test-plan)

• Designed thermal shutdown circuit (from scratch) and testmux

• Eval/debug of silicon; metal fixes

• Key role in top-level simulations

Project Detail – GPS PMIC

Buck regulators (Vin 1.8-6.5V, Vout 0.7-3.3V, 6-bit programmable, 2A max load, 2.2 uH, 4.7-10 uF) were voltage-mode; efficiency ~87%; considerable digital content

First-pass silicon did not permit system oscillator to be trimmed because clock was not mux'ed out by the digital. Shipping of untrimmed units during sampling phase led to undesirable/upredictable behavior in the field

Buck regulators would delay 384 clock cycles (to allow internal bias nodes to be ready) before beginning switching activity. This deterministic delay was used to trim the system oscillator. This workaround permitted about 800k units to be shipped before 1P1 silicon was available

Project Detail – DSC PMIC

• 1Q-2Q 2007

• Cost-reduced (area efficient) version of a production PMIC for digital still cameras

• Chip contained boost regulator, 1 buck regulator, backlight LED boost regulator, LDOs, 2 buck-boost regulators

• Taped out as a test-chip. Only partial silicon evaluation

• Worked on buck-boost and top-level simulations

Project Detail – DSC PMIC

• About 5% of units had an issue with the buck-boost regulators. In light-load, inductor current would randomly go negative and Vout would drop and system would reset

• It was found that these units had the PFM-PWM transition point set very high. At test, units would be trimmed to set this transition to a value where the problem was not observed. Some yield loss remained on units that could not be trimmed acceptably

• Was also found that the variation (across units) of this transition (pre trim) was very wide

• Traced this variation to the Ios of the current-sense amp; reduced it in the new design

Project Detail – Scanner AFE

• 3Q 2006 – 1Q 2007

• Aggressive schedule; process was new to the team

• Chip contained 240 MHz osc for USB2 data rate, 12 bit ADC (2 MSPS), CDS-PGA interface to CMOS/CCD image sensor, offset DAC, LVDS, LED driver

• System used bare die – no trim available (issues with both EEPROM and poly fuse)

• Designed bandgap reference, reference buffer, 1.5V buffer, pipeline ADC differential reference buffer and CDS-PGA amp (scaled down for use as pipeline-ADC residue amp)


• Bandgap reference needed to be fully-isolated. PNP based topology exists, but better curvature performance available from an NPN bandgap

• Known fully-isolated NPN bandgap has non-trivial startup – difficult to sense the NPN current – and some corners showed oscillations during startup

• Developed new topology to meet requirements

• Designed for statistical variation (no trim)

• Bandgap noise filtered with RC filter; buffers designed for noise; total integrated noise at ADC reference output (1 uF caps) was 50 uV rms


• CDS-PGA amp was a two-stage design due to required voltage swing

• Design used cascode compensation and barely met settling requirements but exceeded THD requirements

• However, power consumption was excessive

• Redesigned prior to tapeout with Miller compensation by another resource


• Considerable wafer probing effort after first silicon to determine cause of bandgap not powering up

• Traced to a software error in mask generation. N-wells of PMOS devices shared with NPN collectors resulted in NLDD being applied – which resulted in very large Vth for these PMOS devices in these few circuits

• Re-tapout with corrected masks gave working silicon

• 100% revenue share for TI during year 1 after Wolfsen failed to deliver on time

Project Detail – Display PMIC

• 4Q 2005 – 2Q 2006; targeted at laser projection display system; co-designed with customer

• Chip contained switching regs, linear regs, charge pump, piezo driver, fan controller, RF section – video interface, XTAL oscillator, low-offset amps, laser drivers

• Designed several blocks from scratch and also did a feasibility study to ensure laser driver speed could be attained in the 0.6 um process

• First silicon worked, no issues in blocks designed personally. However, project was put on hold after customer was acquired by Motorola

Project Detail – Display PMIC• Designed bandgap reference to be low noise (~600 uA IQQ);

bandgap buffer, reference current generator, 2 linear regulators (1.2V and 3.0V) – also low noise

• Linear regs used internal zero compensation. Error amps were bipolar input-pair to eliminate flicker noise

• Designed <50 uV Vos autozero opamp for use in green laser temp stabilization loop using topology similar to OPA335. Starting point was OPA333 which uses chopper stabilization. Made substantial modifications to input and output stages, but topology was preserved

• Also designed supply-monitor block (from scratch), testmux (reuse, customization), I/O buffers (reuse, verification) and laid out IREF block

Project Detail – ADI Internship

• Targeted <10 uV Vos, achieved ~20 uV worst case, classic autozero scheme. Explored a scheme intended to give a low-noise AZ amp, later proved invalid

• Designed a temperature insensitive current reference by imposing VPTAT upon a resistor with 3333 ppm TC made using the POLY and NWELL resistors

• Designed a simple low-power (~20 uA) PLL clock-multiplier using the Maneatis scheme without which the chip would have uncoupled oscillators

• All blocks at schematic only – no silicon

Georgia Tech Research

• Goal was to design an optoelectronic receiver with acceptable BER at 10 Gbps with 10 uA p2p signal current from a photodiode using a 0.18 um CMOS process without inductors

• Concept was to use resistive degeneration to get an inductive peaking effect to improve bandwidth

• Individually laid out and submitted one 0.5 um CMOS chip with legacy receivers and demonstrated 400 Mbps performance in the lab

• Receiver was a transimpedance amp (TIA) which converted the single-ended photodiode current into a differential voltage, followed by low-gain high-BW stages and a 50 ohm driver (to the BERT)

Project Detail - TFP7403• Flat panel display timing device (3Q 2000 – 1Q 2001)

• Designed input and output PLLs (VCO/PFD/etc already available) per the Maneatis scheme (JSSC '96)

• Charge-pump current is ratioed to the VCO cell current and loop-filter buffer size is ratioed to the VCO cell size. This gives loop bandwidth to reference frequency to be independent of process and also constant damping factor

• Design involves picking these ratios for optimal bandwidth and acceptable phase margin

• Macromodeling done to speed up simulations, linear modeling done to estimate phase margin, added additional passive filtering to reduce reference spur

• Later, ported VCO to a new process

Project Detail - AIC12

• 16-bit 26 KSPS Delta Sigma Analog Interface Circuit in 0.35 um CMOS (1H 2000)

• Delta Sigma modulator output was 4 level. DAC + SCF was a fully-differential single-opamp ckt that implemented a simple switched-cap filter and DAC (split sampling caps)

• Also worked on an 8 ohm speaker driver during early phase of the project

Project Detail – Recyclic ADC

• Designed a 14-bit recyclic ADC reusing components from the TLV1562 (1.5bit/stg pipeline ADC)

• Changed gain-stage to Steven-Lewis style (input sampled on feedback cap also – so that feedback factor is 1/2 instead of 1/3 – higher loop bandwidth

• Designed error-correction logic

• Recyclic ADC is basically a single-stage pipeline ADC – so the latency is similar to a SAR

Project Detail – DSC AFE TLV977

• Designed (reuse, modifications for speed) resistor-string DACs for use in offset cancellation loop

• Did complete hookup of the 12-bit 18 MSPS pipeline ADC. HDL coding, synthesis and autolayout of the error-correction logic

• Modeled the ADC in VHDL and later modeled the full chip in VHDL to verify top-level hookup

• TI-India, 1999

Project Detail – THS8133 and Remix

• THS8133 – 10-bit 80 MSPS current-steering DAC. Coded the digital interface in VHDL (1998)

• Remix – 10-bit 6 MSPS SAR ADC testchip – coded/synthesized/autolayed out digital interface (1998)

Skills/Strengths

• Analog circuit design

• Analog sub-system (PLL/ADC/DAC/etc) design for moderate performance

• Analysis – circuits and root-cause

• Tradeoffs – area/power/speed/noise

• Scripting – perl/shell/skill

• Veriloga modeling / macro-modeling

• Tools

Patents

• Fully-isolated bandgap reference

• Method of sensing PTAT current in a fully-isolated bandgap-reference

• Fully-isolated temperature-threshold detector

• Low-voltage bias generator not requiring compensation capacitance (pending)

• Low-voltage, low-area bandgap-referenced power-on-reset circuit

Backup


• In October, focus shifted completely to debug and RTP of current device; new development put on hold

• LDO failure at high-temp was due to use of a 1 nA current to generate a delay in transitioning to full-strength during startup to limit inrush current. 1 nA current was vulnerable to leakage

• Conceived and implemented bench tests to collect data supporting the hypothesis and give the customer confidence in the root-cause explanation of this issue which was gating RTP. Interfaced with modeling team to get true worst-case model for leakage. Also developed the metal fix for the issue


• Worked on LDO transient issue debug (RCA of large spike on LDO_DIG output when waking up from sleep mode) – explained by asymmetric charge/discharge strength of stage driving pass device. Not a showstopper; addressed on new device design with a parallel fast-mode that detected and corrected the situation with about 1 uA additional IQQ overhead

• For new design, also achieved the strength transition delay without use of a tiny current and without area overhead by eliminating a binary-2-therm trim decoder


• Last minute issue gating RTP was a low-rate (0.2% of units) powerup failure in the system

• Aggravated at low temp (failing unit would pass at high temp). As temp increased, unit would pass and DIG LDO voltage ramp rate was not deterministic

• Explained observations based on fact that hysteresis was not used in implementing the powerup threshold (no noise immunity) (comparison of bandgap with Vbe)

• Statistics of the fail explained by showing the level-shifter powerup value (with VDDIN absent (DIG LDO)) was HI only 80% of the time (mismatch), and required trim input to the bandgap reference for the fail


• Developed spreadsheets that helped RX architecture definition

• Extensive perl/shell scripting developed to mimic a test prescribed by the customer to generate a large number of datasets from transient simulation (with transient noise) to validate architecture

• Identified opportunities to save power in the design by improving RX components (VMID buffer in the SAR ADC, antialias filter amp (class AB o/p stage systematic mismatch))

• Also found a way to save area in the HF Osc trim DAC by sharing nwells in the cross-coupled latches

Fully-Isolated Design

Definition – no node in the circuit (with the exception of supply and GND) is permitted to touch the (noisy) substrate

When the process provides a vertical NPN with high beta, usually the collector is an n-well – which touches the substrate

Therefore, a bandgap reference topology such as the Brokaw bandgap, in which the collectors of the NPNs are used in the circuit, is not permitted

Trajectory at Texas Instruments (Dallas)

Hired into Custom Mixed Signal group in Oct 2005 Moved to High Performance Analog (HPA) group in

July 2008 (project had been outsourced to HPA by ASIC)

DSPS-R&D begins taking over project in November 2008

Restructuring in Dec 2008 led to resources working on the project being moved to ASIC

Majority of project team affected by RIF in 2009

Ananthprofilepln

Documents

new board

new chip

large compcap

texas instruments india

sar adc initiative

pipeline adc

small comp cap amd millercomp

current reference