a programmable multi-channel sub-threshold fir filter for a body sensor node

24
Robu st Low Powe r VLSI Robust Low Power VLSI A Programmable Multi-Channel Sub-Threshold FIR Filter for a Body Sensor Node Alicia Klinefelter Dept. of Electrical Engineering, University of Virginia January 16, 2012

Upload: denim

Post on 22-Feb-2016

28 views

Category:

Documents


0 download

DESCRIPTION

A Programmable Multi-Channel Sub-Threshold FIR Filter for a Body Sensor Node. Alicia Klinefelter Dept. of Electrical Engineering, University of Virginia January 16, 2012. Motivation. Wireless body sensor nodes (BSN) well-suited for sub-threshold - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

RobustLowPowerVLSI

A Programmable Multi-Channel Sub-Threshold FIR Filter for a Body Sensor NodeAlicia KlinefelterDept. of Electrical Engineering, University of VirginiaJanuary 16, 2012

Page 2: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

2

Motivation Wireless body sensor nodes (BSN) well-suited

for sub-threshold Accelerators more energy efficient than MCU

No multiplier on MCU Filtering operation frequently used

Application: EEG signal power extracted from multiple frequency bands Prior work used analog multi-channel FIR for energy extraction [4]

A need for filtering flexibility Portability

Page 3: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

3

Outline Design Overview

Context: Full chip FIR Overview

Filter Decisions and Tradeoffs Filter topologies Filter and Channel Design Leakage Reduction

Filter Features Results

Design Comparison

Future Work

Page 4: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

4

BSN Overview 19µW chip including

analog front-end (AFE), memory, digital processing, power management and TX

Ultra-low power: Batteryless Harvested energy

FIR part of flexible data path.

BSN Node Chip Micrograph [3]

BSN Node Datapath Flexibility [3]

Page 5: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

5

FIR Overview Configurable/Programmable

Number of taps Number of filters Coefficients

Four input and processing channels Synthesized and fabricated in a 130nm

technology using the Cadence design flow: Verilog RC Compiler Encounter Place and Route Virtuoso

Operates down to 300mV at 8kHz Employs clock and power gating for energy

savings

Page 6: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

6

Outline Design Overview

Context: Full chip FIR Overview

Filter Decisions and Tradeoffs Filter topologies Filter and Channel Design Leakage Reduction

Filter Features Results

Design Comparison

Future Work

Page 7: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

7

z-1

b0

x[n]

+

z-1

b1

+

z-1

a0

+

z-1

a1

+ y[n]

Architectures for Low Power: IIRs

Infinite impulse response (IIR): fewer taps, sharper cutoff

Non-linear phase tolerable for application Instability a big problem

𝑯 (𝒛 )𝑰𝑰𝑹=∑𝒌=𝟎

𝑴𝒃𝒌 𝒛−𝒌

𝟏+∑𝒌=𝟏

𝑵𝒂𝒌 𝒛−𝒌

Page 8: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

8

IIR: Instability

Desired cutoff results in poles near unit circle

Page 9: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

9

Architectures for Low Power: FIR

Direct form FIR More coefficients to achieve desired cutoff

Symmetric coefficients

No feedback No stability problems

𝑯 (𝒛 )𝑭𝑰𝑹=∑𝒌=𝟎

𝑴𝒃𝒌𝒛−𝒌

z-1 z-1 z-1 z-1

b0 b1 b2 b3 bn

x[n]

+ + + +...

...

y[n]

Page 10: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

10

Channel Design

Resource-shared architecture [2] 1 adder, 1 multiplier per channel

1 tap computed per clock cycle 195-781 fast clock cycles per

sample clock period

Channel control logic Maintains channel state Clock gating control

x

+ACCUMREGs

. . .

. . .

CHANNEL CONTROL

LOGIC

x[n]

y[n]

x[n-1] x[n-k]

b0

b1

bk

b0

b0x[n]

x[n]

0

y[n] = b0x[n]

x[n-1]

b1

b1x[n-1]

b0x[n]

y[n] = b0x[n]+b1x[n-1]

fast clock . . .

x[n-k]

bk bkx[n-k]

b0x[n]+…+bk-1x[n-k-1]

y[n] = b0x[n]+…+bkx[n-k]

sample clock

Page 11: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

11

FIR Block Diagram

x

+ACCUMREGs

. . .

. . .

CHANNEL CONTROL

LOGICCHANNEL

1

CHANNEL 2

CHANNEL 3

CHANNEL 4

x[n]

y[n]

x[n-1] x[n-k]

b0

b1

bk

Clock Divider

x[n]

out1

out2

out3

out4

Channel Sync Logic

System Clock

Sample Clock

Sampled Data (x[n])

Channel Computation

t

old_sample new_sample

result available

Sample Clock

System Clock

y[n]1,2,3,4intermediate results

Page 12: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

12

Sleep Mode Power Savings

Power gating For when block is not on the

datapath Simulated power gated

channels

Clock gating Many fast clock cycles not

used per sample period Clock gate all channels after

result computed or block is off

FIR

VDDlow VDDhigh

Enlow EnhighW1/L W2/L

W2>W1

Page 13: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

13

Outline Design Overview

Context: Full chip FIR Overview

Filter Decisions and Tradeoffs Filter topologies Filter and Channel Design Leakage Reduction

Filter Flexibilty Results

Design Comparison

Future Work

Page 14: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

14

Features: Taps Selection

Prior works has 8-14 taps E/sample increases with more taps Throughput still met with more clock cycles

Page 15: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

15

Features: Number of Taps

Programmable number of taps Half taps mode (15 taps) for less accurate results Full taps (30 taps) for a more accurate result Can use adder on chip’s CPU to create 60 tap filter

Programmable number of filters

CHANNEL 0

CHANNEL 1

CPU

+

Channel 1 Data Ready

60 tap result

x[0] à x[29]

x[30] à x[59]

Page 16: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

16

Outline Design Overview

Context: Full chip FIR Overview

Filter Decisions and Tradeoffs Filter topologies Filter and Channel Design Leakage Reduction

Filter Features Results

Design Comparison

Future Work

Page 17: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

17

Results: Frequency Response

(a) (b)

(c) (d)• Measured frequency response for varying tap lengths (a) 18-12Hz (b) 18-26Hz

(c) 30-50Hz (d) 70-100Hz

Page 18: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

18

Measured Results: ED Curve

350mV, 28kHz

350mV, 22kHz

Page 19: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

19

Measured Results: EEG Filtering

time(s)

Volta

ge (V

)(a)

(b)

(c)

(d)

(e)

• Filtering of EEG data set. (a) Original signal sampled at 250Hz (b) filtered at 8-12Hz (c) filtered at 18-26Hz (d) filtered at 30-50Hz (e) filtered at 70-100Hz

*data from [1]

f (Hz)

|Y(f)

|

Page 20: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

20

Design ComparisonThis Work [5] [6] [4]

Type 30-tap, 8-bit

8-tap, 8-bit

14-tap, 8-bit

4th order analog

Channels 4 1 1 4

Programmable a r r aTechnology 0.13μm 0.13μm 0.13μm 0.13μm

Supply 0.4V 0.2V 0.27V 1.2V

Frequency 100kHz 12kHz 20MHz 20kHz

Power 118nW 114nW 310μW 780nW

Energy 1.18pJ 9.5pJ 15.57pJ 39pJ

FOM* 0.61 18.55 17.37 N/A

*FIR FOM: power(nW)/frequency(MHz)/# of taps/input bit length/coefficient bit length

Page 21: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

21

Outline Design Overview

Context: Full chip FIR Overview

Filter Decisions and Tradeoffs Filter topologies Filter and Channel Design Leakage Reduction

Filter Features Results

Design Comparison

Future Work

Page 22: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

22

Future Work Fine-grained power gating

analysis Programmable number of

taps: any number Increased Channel flexibility

Process all 4 channels in parallel

Dynamic programming options Reduce register overhead

through use of latches or data memory

CH0

CH2 CH3

CH1

Page 23: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

23

References1. R. Leeb, , C. Brunner, G. R. Muller-Putz, A. Schlogl, and G. Pfurtscheller. “BCI

Competition 2008 - Graz data set B 1”. Institute for Knowledge Discovery, Graz University of Technology, Austria, Institute for Human-Computer Interfaces, Graz University of Technology, Austria.

2. Davis, W.R. , et al., "A design environment for high throughput, low power dedicated signal processing systems," Custom Integrated Circuits, 2001, IEEE Conference on , 2001.

3. Fan Zhang, et al., "A Batteryless 19μW MICS/ISM-Band Energy Harvesting Body Area Sensor Node SoC," International Solid-State Circuits Conference (ISSCC), 2012 IEEE , Feb. 2012.

4. Fan Zhang, et al., "A low-power multi-band ECoG/EEG interface IC, “Custom Integrated Circuits Conference (CICC), 2010 IEEE , Sept. 2010.

5. Myeong-Eun Hwang, et al., “A 85mV 40nW Process-Tolerant Subthreshold 8x8 FIR Filter in 130nm Technology," VLSI Circuits, 2007 IEEE Symposium on , June 2007.

6. Wei-Hsiang Ma, et al., "187 MHz Subthreshold-Supply Charge-Recovery FIR," Solid-State Circuits, IEEE Journal of , April 2010.

Page 24: A  Programmable Multi-Channel Sub-Threshold FIR Filter for a Body  Sensor  Node

RobustLowPowerVLSI

24

Thank You

Questions?