Asynchronous Design for Analogue Electronics Alex Yakovlev - - PowerPoint PPT Presentation

asynchronous design for analogue electronics
SMART_READER_LITE
LIVE PREVIEW

Asynchronous Design for Analogue Electronics Alex Yakovlev - - PowerPoint PPT Presentation

Asynchronous Design for Analogue Electronics Alex Yakovlev Motivation: A4A scope IP core IP core conventional (big digital) (big digital) RTL synthesis ADC sensor sensor DAC level shifters/ analogue synchronisers power power


slide-1
SLIDE 1

Asynchronous Design for Analogue Electronics

Alex Yakovlev

slide-2
SLIDE 2

Motivation: A4A scope

2 / 31

conventional RTL synthesis ad hoc requires formalisation and design automation analogue components

sensor synchronisers level shifters/ sensor power converter power converter analogue digital

time/energy infrastructure

A4A scope ADC DAC

IP core IP core control for analogue layer (little digital) manual design (big digital) (big digital)

  • Analogue and digital electronics are becoming more intertwined
  • Analogue domain becomes more complex and itself needs

digital control

slide-3
SLIDE 3

Motivation: Power electronics context

3 / 31

  • Efficient implementation of power converters is paramount
  • Extending the battery life of mobile gadgets
  • Reducing the energy bill for PCs and data centres

(5% and 3% of global electricity production respectively)

  • Need for responsive and reliable control circuitry - little digital
  • Millions of control decisions per second for years
  • An incorrect decision may permanently damage the circuit
  • Poor EDA support
  • Synthesis is optimised for data processing - big digital
  • Ad hoc solutions are prone to errors and cannot be verified
slide-4
SLIDE 4

Basic buck converter: Schematic

4 / 31 control

Th_nmos Th_pmos

buck

V_ref V_0

R_load PMOS NMOS

I_max gp_ack

  • c

uv zc gn_ack gp gn

  • ver-current (oc)

under-voltage (uv) zero-crossing (zc)

  • In the textbook buck a diode is used instead of NMOS transistor
slide-5
SLIDE 5

Basic buck converter: Informal specification

5 / 31

UV UV OC I_max

current

no ZC late ZC

OC PMIN

early ZC

PMOS OFF ZC PMOS OFF NMOS ON NMOS OFF PMIN ZC NMOS OFF PMOS ON NMOS OFF N M O S O N P M O S O F F NMOS ON PMOS OFF PMOS ON UV OC

time

NMOS OFF PMOS ON NMIN NMIN PMIN

  • no ZC – under-voltage without zero-crossing
  • late ZC – under-voltage before zero-crossing
  • early ZC – under-voltage after zero-crossing
slide-6
SLIDE 6

Multiphase buck converter: Schematic

6 / 31

Th_pmos I_max

NMOS[1]

gn1 gn_ack1 Th_nmos zc1 V_0

PMOS[N]

gp_ackN

  • cN

hl

zero-crossing (zc)

Th_nmos V_0

NMOS[N]

under-voltage (uv)

V_ref gnN

PMOS[1]

gp1 Th_pmos gp_ack1 I_max

  • ver-current (oc)
  • c1

zcN uv gn_ackN gpN V_min

high-load (hl)

control buck

R_load

slide-7
SLIDE 7

Multiphase buck converter: Informal specification

7 / 31

  • Normal mode
  • Phases are activated sequentially
  • Phases may overlap
  • High-load mode
  • All phases are activated simultaneously
  • Benefits
  • Faster reaction to the power demand
  • Heat dissipation from a larger area
  • Decreased ripple of the output voltage
  • Smaller transistors and coils
slide-8
SLIDE 8

Synchronous design

8 / 31

  • Two clocks: phase activation (~5MHz) and sampling (~100MHz)

Easy to design (RTL synthesis flow) Response time is of the order of clock period Power consumed even when idle Non-negligible probability of a synchronisation failure

  • Manual ad hoc design to alleviate the disadvantages

Verification by exhaustive simulation

slide-9
SLIDE 9

Asynchronous design

9 / 31

  • Event-driven control decisions

Prompt response (a delay of few gates) No dynamic power consumption when the buck is inactive Other well known advantages Insufficient methodology and tool support

  • Our goals
  • Formal specification of power control behaviour
  • Reuse of existing synthesis methods
  • Formal verification of the obtained circuits
  • Demonstrate new advantages for power regulation

(power efficiency, smaller coils, ripple and transient response)

slide-10
SLIDE 10

High-level architecture: Token ring

10 / 31

  • No need for phase activation clock
slide-11
SLIDE 11

High-level architecture: Charging stage

11 / 31

slide-12
SLIDE 12

A2A components

12 / 31

  • Interface analogue world of “dirty” signals
  • Provide hazard-free “sanitised” digital signals for asynchronous control
  • Library of A2A components
  • WAIT / WAIT0

– wait for stable 1 / 0 on analogue input and latch it until explicit release signal

  • RWAIT / RWAIT0 – WAIT element with a possibility to

persistently cancel the waiting request

  • WAIT01 / WAIT10 – wait for a rising / falling edge
  • WAITX / WAITX2 – wait for one of analogue signals

in a mutually exclusive way using 4-phase / 2-phase control signalling

  • SAMPLE

– sample the state of analogue signal

slide-13
SLIDE 13

A2A components: WAIT element

13 / 31

  • STG specification

Read-consume conflict between output and input.

  • ME-based solution
  • Gate-level implementation

can be removed

slide-14
SLIDE 14

Synthesis flow

14 / 31

  • Manual decomposition of the system into modules
  • To create formal specification from informal requirements

(feedback loop with engineers)

  • To simplify specification and synthesis
  • Some modules are reusable
  • Some modules (A2A components, Opportunistic Merge)

are potential standard components

  • Each component is specified using STGs
  • Automatic synthesis into speed-independent circuits

(arbitrary gate delays and some forks must be isochronic)

slide-15
SLIDE 15

Formal verification

15 / 31

  • STG verification
  • All standard speed-independence properties

(consistency, output-persistency, complete state coding)

  • PMOS and NMOS are never ON simultaneously

(to prevent from short circuit)

  • Some timers are used in a mutually exclusive

way and can be shared

  • Circuit verification
  • Conforms to the environment
  • Deadlock-free and hazard-free under the given environment
slide-16
SLIDE 16

Tool support: WORKCRAFT

16 / 31

  • Framework for interpreted graph models (STGs, circuits, FSMs,

dataflow structures, etc.)

  • Interoperability between models
  • Elaborated GUI
  • Includes many backend tools
  • PETRIFY – STG and circuit synthesis, BDD-based
  • PUNF – STG unfolder
  • MPSAT – unfolding-based verification and synthesis
  • PCOMP – parallel composition of STGs
slide-17
SLIDE 17

Tool support: WORKCRAFT

17 / 31

slide-18
SLIDE 18

Simulation results

18 / 31

  • Verilog-A model of the 3-phase buck
  • Control implemented in TSMC 90nm
  • AMS simulation in CADENCE NC-VERILOG
  • Synchronous design
  • Phase activation clock – 5 MHz
  • Clocked FSM-based control – 100 MHz
  • Sampling and synchronisation
  • Asynchronous design
  • Phase activation - token ring with 200 ns timer (= 5 MHz)
  • Event-driven control (input-output mode)
  • Waiting rather than sampling (A2A components)
slide-19
SLIDE 19

Simulation results

19 / 31

synchronous asynchronous

slide-20
SLIDE 20

AMS Trends & Challenges

20 / 31

  • Key drivers
  • Internet of Things
  • Mobile computing
  • Automotive electronics
  • Trends
  • Technology scaling
  • Multiple power and time domains
  • Analog and digital integration
  • Challenges
  • Tighter reliability margins
  • Concurrent analog and digital analysis
  • Short development cycle

Based on slide from DAC-2014 by ANSYS

slide-21
SLIDE 21

What this means for AMS?

21 / 31

  • Achieving better verification of analog and digital blocks
  • Verifying the increasing amount of digital logic in

analog designs

  • Creating a higher level of abstraction for analog and

mixed signal blocks

  • Automating the manual custom design steps
  • Adopting circuit analytics that tell why and where the

circuit is failing to perform

Based on slide from ISQED-2013 by Mentor Graphics

slide-22
SLIDE 22

Advanced AMS design flow

22 / 31

slide-23
SLIDE 23

AMS flow: Tool support

23 / 31

  • WORKCRAFT – synthesis and verification of async. circuits
  • LEMA – modelling and verification of AMS circuits
  • Model generation from simulation traces
  • Property expression and checking
slide-24
SLIDE 24

AMS flow: Basic buck

24 / 31

control

Th_nmos Th_pmos

buck

V_ref

R_load PMOS NMOS

I_max gp_ack

  • c

uv gn_ack gp gn

  • ver-current (oc)

under-voltage (uv)

slide-25
SLIDE 25

AMS flow: Model generation example

25 / 31

1

gp

1

gp_ack

5 10 1180 1185 1190 1195 1200 1205 1210 1215 1220 1225 1230 1235

PMOS voltage V ns

gp_ack

✂ ✄ ☎ ❣ ✆ ❂
✂ ✄ ☎ ❣ ✆ ✝ ❣ ✁ ✞ ☎ ❂ ✟ ✠ ✠ ✠ ✠ ❣ ✆ ✝ ❣ ✁ ✞ ☎ ✝ ✡ ✁ ✞ ☎ ❂ ✠

Initial conditions:

☛ ☞ ✌ ☛ ✍ ✎ ✏ ✌ ✑ ✍ ✎

e

☛ ☞ ✌ ☛ ✍ ✎ ✏ ✌ ✑ ✍ ✎

e

☛ ☞ ✌ ☛ ✍ ✎ ✏ ✌ ✑ ✍ ✎

e

☛ ☞ ✌ ☛ ✍ ✎ ✏ ✌ ✑ ✍ ✎

e

☛ ☞ ✌ ☛ ✍ ✎ ✏ ✌ ✑ ✍ ✎

e

☛ ☞ ✌ ☛ ✍ ✎ ✏ ✌ ✑ ✍ ✎

e

slide-26
SLIDE 26

AMS flow: Optimised specification

26 / 31

  • Concurrency reduction
  • Scenario elimination
slide-27
SLIDE 27

ADC: Sampling schemes

27 / 31

  • Synchronous

ADC

Ts

  • Asynchronous

AADC

  • A. Ogweno, P

. Degenaar, V. Khomenko and A. Yakovlev: “A fixed window level crossing ADC with activity dependent power dissipation”, accepted for NEWCAS-2016.

slide-28
SLIDE 28

ADC: Asynchronous design

28 / 31

Vin

Vramp rise Vpulse fall Vchange

refn refp refm

t2 t3 t4 t5 t6 t1

slide-29
SLIDE 29

ADC: Specification and implementation

29 / 31

  • STG specification
  • Speed-independent implementation
  • utn-
  • utp-

req+ req+

  • utp+
  • utn+

req- req- ack- ack+ ack- p1 p2 p3

ack ack req

  • utn
  • utp

req reset ramp_en Vpulse

slide-30
SLIDE 30

Conclusions

30 / 31

  • Fully asynchronous design of multiphase buck controller
  • Quick response time: few gate delays, all mutexes are
  • utside the critical path
  • Reliable: no synchronisation failures
  • Design flow is automated to large extent
  • Automatic logic synthesis
  • Formal verification at the STG and circuit levels
  • Library of A2A components
  • Vision for an advanced AMS design flow
slide-31
SLIDE 31

Acknowledgements

31 / 31

  • A4A team
  • Newcastle University:

Vladimir Dubikhin, Victor Khomenko, Andrey Mokhov, Danil Sokolov, Prof. Alex Yakovlev

  • Dialog Semiconductor: David Lloyd
  • University of Utah:
  • Prof. Chris Myers
  • EPSRC for funding A4A project (EP/L025507/1)
  • Design automation
  • WORKCRAFT – http://workcraft.org/
  • LEMA – http://www.async.ece.utah.edu/LEMA
  • Recent publications
  • D. Lloyd, R. Illman: “Scan insertion and ATPG for C-gate based

asynchronous designs”, SNUG- 2014.

  • D. Sokolov, et. al: “Design and verification of speed-independent buck

controller”, ASYNC-2015.

  • V. Dubikhin, et. al: “Design of mixed-signal systems with asynchronous

control”, IEEE Design & Test (to appear).