ISSCC Tutorial 2011

Interfacing Silicon to the

Human Body
Timothy Denison, PhD

Director of Neuroengineering
Technical Fellow, Medtronic

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 ACKNOWLEDGEMENTS FIRST!

• Sensing and Algorithm Work, General Systems:

Chris Van Hoff and Firat Yazicioglu
(significant slide/thought contributions)
• Chemical Sensing in the Brain:
Pedram Mohseni
• Actuation and Retinal Implants:
Albrecht Rothermel, M Ortmanns
• Low Power Design Principles:
Rahul Sarpeshkar
• Digital Signal Processing:
Joyce Kwong, Anantha Chandrakasan

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Interfacing “Electrical Signals” to the body is
a time-honored tradition: Scribonius Largus

“… Headache, even if it is chronic and unbearable, is taken

away and remedied forever by a live torpedo [fish] placed on
the spot which is in pain, until the pain ceases.”

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Interfacing Silicon to the Body, A Systems Approach

Tutorial and Architectural Overview

Recurring Themes through Tutorial

• Management of Energy
• Management of Information
• Management of Interfaces

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Chapter 1
“Actuation” Systems and Interfaces

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 The Power of “Actuation”
The patient has an activity tremor  tremor while moving

Pre-Op

Double click box to play video

Electrical stimulation …
helps address a circuit malfunction in the nervous system

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Presenting Problem: Neuromodulation Design

System includes:  Energy: sources and management

 Information: biophysics, circuits, coding
 Interfaces 1: electrodes (Safety, Impedance)
 Interfaces 2: packaging (Hermetic, Bias)
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Deriving Specs 1: Patterns for Neural Coding
Engineering Design and Architecture

Biological Anatomy and Physiology

Band-pass filters:

LOW HIGH

Frequency (Hz)
7400
625
250
375
500

SELECT BANDS WITH CONVERT ACOUSTIC

HIGHEST AMPLITUDES TO ELECTRIC LEVELS

APICAL BASAL
Electrode array

Interfaces Mesh the Systems Together

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Deriving Specs 2:
Waveform Parameters for Modulating Activity

Generalized stimulation pulse train. Q(c) is cathodic charge flow out of the tissue; t(c) is the duration of
the cathodic charge flow; t(d) is the interval between cathodic and anodic phases; Q(a) is the anodic
charge flow into the tissue; t(a) is the duration of the anodic charge flow; F is the frequency of charge
delivery. The solid line for Q(a) indicates “active” anodic charge recovery versus the “passive” charge
recovery illustrated with the dotted line. Adapted from Troyk, P.R., Cogan, S.F. “Sensory Neural
Prostheses.” Neural Engineering. Ed. He, B. New York: Kluwer, 2005.

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Primer: Physiology of Electrical Stimulation

Stimulation
Electrode

Na+ K+
External K+
Medium
Voltage

Open
Channel
Na+
Voltage

K+
Current

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Example: Neural Activation Functions
Strength-Duration Curve

Engineering
Abstraction

Data from [Rothermel2008]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Typical Stimulation Requirements
for Neuro and Cardiac Stimulators

Neurostimulator Cardiac Pulse Generator

Frequency 1 – 250+ Hz 0.5 Hz - 3.5 Hz

Cathodic / anodic 250 µA – 15 mA 1 mA – 15 mA
current (peak)
Duration of 30 µs – 2 ms 120 µs – 1.5 ms
cathodic current
Number of 16 or more 2 (atrial / ventricular)
electrodes
Sensed signal > 1 µV, 12-40 Hz (beta band) > 150 µV, 13-30 Hz (atrial
amplitude depolarization)
Battery type Secondary or Primary Primary
Nominal 63 µA (3 mA, 210 µs, 100 Hz) 5.6 µA (7 mA, 400 µs, 1 Hz),
stimulation current 100% A+V pacing
Example ISO 14708-3 (INS), EN EN 45502-2-1, ISO 14708-2
regulatory 60118-13 (Cochlear
standard Implant)
differences

100μW to 1mW of energy 10μW of energy

Prosthesis > 10x this amount

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Interfaces: How Do We Activate Tissue?
Tissue Conduction is Ionic, Electronics is …

Hypothesized Double Layer @ Charged Exchange at the electrode (Non-polarized)

Electrode-Tissue Interface Electrode-Tissue Interface

Hypothesized Structure
Of Double Layer: Polarization Impacts:
Region “A”: Surface hydration. • Distortion of evoked response.
Region “B”: Loosely held • Energy losses due to polarization Z
hydration layer with hydrated Cations.
Region “C”: Bulk Solution. • Safe charge transfer

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Some Interface Constraints: Multiple standards

• 30 μC/cm2 charge density per phase limit. Long

history with FDA. Favorable standard for large
electrodes.

• K = 1.5 to K = 1.7 from the Shannon relation:

log10(Q) + log10(QD) = K. Favorable standard for
small electrodes. Q = Net charge/phase
(μCoulombs), QD = Charge density (μCoulombs/cm2)

• Most work based on acute animal stimulation tests (7

hours). This is limit for stimulating an anesthetized
animal.

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 McCreery: intracortical and cortical stimulation
of cat parietal cortex safe to K=1.7
(Shannon equation with 7 hours of stimulation)

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 System-Level: Generating Stimulation Signal
[Denison2010c]

Energy

 “Unipolar” mode: the case of the INS as the current sink or source.
 “Bipolar” mode: the current sink and source in distributed electrodes.
 Capacitors used to block DC current (most of the time…)
 Design Challenge: Active circuitry to prevent charge accumulation and
destructive faradic current across the electrode / tissue interface

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Example of a Stimulation Circuit (“Stim Engine”)
VHigh VHigh
ΦC1 RRef RO
ΦC2 ΦC1 ΦO
A B
C ΦC1
ΦC512
ΦO
ΦO ΦC1 ISource1
IBias
ISource2
IIN IOut
+ ISource3
+ IRef VHigh
-
-
ΦC512 RO
ΦO
C ΦC5
D7 D7 D0 D0 IRef 12
ΦO
2R 2R 2R ΦC512
IBias I
Source512
R R

[Denison2010a,b,c]
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Conversion from Energy Source to Electrode Drive

Discharge curve for Quallion QL0003I

[Denison2010c]
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Methods of Boosting the Voltage (all used)
 Capacitor stacking [Denison2010a,b,c]
 Advantage: Simple
 Disadvantage:
 Need very low impedance switches
 A large load current as the capacitors are charged
→ very large battery decoupling capacitors and/or low
impedance battery chemistries.
 Capacitors need to be very large to support the droop

 Inductive-based switch-mode power supply

 Advantage: Very efficient with a ferrite core inductor
 Disadvantage:
 Significant volume in the space-constrained environment
 Ferrites lead to unintended movement, thermal injury from RF
induction heating or image artifacts during MRI

 Charge Pump
 Advantages:
 High efficiency
 One large hold capacitor and small pump capacitors
 Adjustable voltage gain
 Feedback can be used to achieve minimum necessary voltage
 Disadvantage: Modest circuit complexity (switch management)

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 X1 to X5 Charge Pump Circuit, and Example

 Highest voltage for

gate control →
Dynamic voltage
comparator and control.
 Multiple voltages for
efficient stimulation

Gain of 1.5 configuration

[Denison2010c]
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 When you get it right…
Essential Tremor Patient Before & After

Comparing Pre & Post-Op

Double click box to play video

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Emergent Application: Retinal Prosthesis
Advanced Stimulation/Actuation Systems
IC requirements
(Info/Energy/Interface)
• Spatial Mapping [1:1]
• Spatial Derivative
• Drive cells  log (I)
• Actuate safely
• Fine pitch (~ 1000 sites)
• Slip under retina

[Rothermel2008]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Concept for Retinal Implant
TiN electrode
Photo diode

Active pixel array

Direct stimulation
test electrodes

[Rothermel2008]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Retinal Implant Supply Architecture
VL
Packaging Concerns drive VOUT
GND
AC-supply concept:
t
VH
Use to minimize corrosion

Output
Rectifier
Drivers

Low Input
Impedance
Amplifiers

[Rothermel2008]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 DC to AC Converter  Apply for Stimulation
Drive characteristics based on physiology
VDDCONST
VL

Analog
input
Electrodes
Address

Current VH
limit VSS
Local stimulation is gradient against background light
Drive is derived from inherent logarithm of optical detector

Replicate
Neural
Code in
Opto-sensor

[Rothermel2008]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 DC to AC Converter  Apply for Stimulation

Recharge Phase on Inverted Supply

VDDCONST
VH

Analog
input
Electrodes
Address

Current VL
limit
VSS

[Rothermel2008]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 But No room for Capacitors!
Solving the Charge Balance Problem
VOUT(t)
1. Load has no resistive path I(t)
to ground:

 Discharge allows to t
remove residual charge

VOUT(t)
I(t)
2. Load has resistive path
to ground:
t
 Charge balance requires
zero voltage integral

[Rothermel2008]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Advanced Concepts for Charge Balance:
Method of Precision via Feedback

Auto-sample the Current Sources Deliver a Bi-phasic Pulse, and Reset

[Sarpeshkar2007]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Advanced Concepts for Charge Balance:
Feedback Loops for Polarization Management

[Ortmanns2009]

• Monitor the polarization, provide feedback

to keep within specified range
• How do we servo this loop?
• How do we measure ground?
• Boluses vs. slow drive…
• Still need to consider details of charge transfer

How do we get free of electrode issues?!

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Evolving Methods: Optogenetic Neuromodulation

Goal: Research Tool

• Better understand networks
• Better understand therapies
• Explore therapy concepts

System Challenges
Biological
•Genetic Transfection
Not •Transfection Efficacy
Activated
Activated Activated Activated •Volume Illumination
Hardware
• Light Source
• Information Flow
• Power Management
• Fluid Environment

Galvanic Neuromodulation Optogenetic Neuromodulation

[Denison2010a,b,c]
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Critical Interface:
Optogenetic “Transconductors”

Channel Modified Channel [3]

[1] Mac Halorhodopsin [4]
Rhodopsin-2 Rhodopsin-2 [2]
Channel Neuronal Neuronal Neuronal Neuronal
Function Excitatory Excitatory Inhibitory Inhibitory
Membrane
Cation Channel Cation Channel Proton Pump Chloride ion pump
Structure
Activation
~450nm (Blue) ~470nm (Blue) ~470nm (Blue) ~580nm (Yellow)
Light λ
Optical Power
8-12mW/mm2 2.4mW/mm2 5.3mW/mm2 10-21mW/mm2
Density
Illumination 535nm (Green) for Illumination Illumination
Deactivation
Cessation 50ms, 2.4mW/mm2 Cessation Cessation
0.01%
Duty-Cycle NA (Continuous) NA (Continuous) NA (Continuous)
(10ms / 5-15sec)
469nm (Blue) Poor Power
Light Source 469nm (Blue) 469nm (Blue)
518nm (Green) Efficiency

Light Sources:
Light Source Power Density Driving Current Efficiency
469nm/Blue 5.3mW/mm2 50mA/4V 3%
518nm/Green 1.4mW/mm2 50mA/3.8V 1%
[1] Nature Neuroscience vol. 8, 1263-1268; PNAS, vol. 100, no. 24, pp. 13940–13945.
[2] Nature Neuroscience, vol. 12, pp. 229–234 (2008).
[3] Nature, vol. 463, pp. 98-102 (2010).
[4] Nature vol. 446, 633-9 (2007).
[Denison2010a,b,c]
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Alternative Therapeutic Stimulation:
Optogenetic Stimulator Prototype

• Chronic & Safe Interface

• Manage Information Flow
• Manage Power Flow

[Denison2010a,b,c]
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Status of Actuation Methods
in the Nervous System (2011)
OCD1 Depression2
775,000

IN DEVELOPMENT
Parkinson's Disease
216,000 Epilepsy2
245,000
COMMERCIAL

Essential Tremor
80,000 Neurodegenerative
Diseases (drug-
Dystonia1 device)3
3,500,000
Severe Spasticity
1,200,000 Migraine
Headache Pain3
Chronic Pain 904,000
1,300,000
Nonopioid
Gastroparesis1 Chronic Pain4
653,000
Overactive Bladder
and etention Fecal Incontinence2
1,800,000 692,000

1. Humanitarian Device Exemption (HDE), 2. Investigational Use Only (IDE), 3. Research,

4. Investigational New Drug Patient #’s = US Net Prevalence (indicated, addressable population)

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Chapter 1 : Summing Up
Actuation Systems and Requirements

• Management of Energy  Information  Interface

• System design and constraints drive IC design
neuroscience, coding, electrode safety, etc
• Actuation methods discussed are stationary…
How do we REACT to dynamic needs and
fluctuations in the body?
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Presenting Problem:
Actuation in Absence of Feedback…

First transistorized pacemaker

was literally a metronome circuit

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Chapter 2
Sensor Systems and Requirements

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 What are the Signals for Detection?
• Cell communication is combination of electrical and chemical activity.
• Electrical Signals  Neural Activity, Muscle, “Derived”
• Chemicals: Neurotransmitters (NT)  learning, memory, mood,
cognition, and neuropathologies (dopamine in Parkinson’s disease
and drug abuse) AND other agents  glucose, etc

~20nm
Scale MATTERS
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Circuit Scaling: Cells  Circuits  Networks

• Origin of signals drives characteristics and requirements

• Choice of scale important in design of an interface!

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Quick Primer: Origins of Bioelectrical Activity

Electrical Models at the

Nano- and Microscale

Amplifiers generally measure

dynamic signals

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Bioelectrical Measurement Principle
(Bi-Polar Measurement)
Measurement
Electrode

Reference AMP VOUT = AV × (V+ − V− )

Electrode

BIAS
Bias Note: Many implantable
Electrode signals are essentially floating

• Subject is biased to a known potential through bias electrode

(not necessarily through direct connection as shown)
• Differential signal between measurement electrodes is
amplified and passed down the chain…
• Common-mode signals are rejected.
Adapted from Yazicioglu

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Cellular Scale: Origins of Bioelectrical Activity
(What we observe with differential system)

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Ensemble Level: Origins of Bioelectrical Activity
(What we observe with differential system)

Typical Neural Signals of Interest

Σ
Sum over network

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Bioelectrical Signals: Amplitude and Frequency Characteristics

EMG
ECG
Amplitude

EEG

ECoG
&
LFP AP

[Webster92]
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Constraints: How Do We Sense These Potentials?
Tissue Conduction is Ionic, Electronics is …

Hypothesized Double Layer @ Charged Exchange at the electrode (Non-polarized)

Electrode-Tissue Interface Electrode-Tissue Interface

Hypothesized Structure
Of Double Layer: Polarization Impacts:
Region “A”: Surface hydration. • Distortion of evoked response.
Region “B”: Loosely held • Energy losses due to polarization Z
hydration layer with hydrated Cations.
Region “C”: Bulk Solution. • Safety considerations

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Power considerations…What are the drivers?

Multiple Considerations

Thermal Dissipation in/on Body

• Few degrees increase max
• ~ 10mW for small implants
• Analysis of Utah Array (Harrison)

Device Longevity
• Therapy ~ 10->100μW ->10mW
• Budget 10% for sensing (?)
• 1-10μW/channel typical for
implantable sensors and algorithm
• Recharge is a burden…

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Application & Constraints Define Specifications
CMRR > 110dB 80dB 60dB
Noise < 0.5μVRMS ~(0.5 -100Hz) 1.0μVRMS (0.1 ->1kHz) 5μVRMS
Inp. Impedance > 100MΩ 5MΩ 100MΩ
DC Headroom > >50mV >50mV >1V
Power/channel < 1mW? 1-10μW (battery) 100μW (thermal)
Leakage Currents < 10uA (CF) 10uA max (usually << 1uA) << 1uA?

EEG/ECG
[IFCN1998]

ECoG

AP
Adapted from Yazicioglu
LFP

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Review: Industry Standard
Instrumentation Amplifier

Lots of power in amplifiers

IN

Lots of power to drive resistors

[Smither1989]
[VanRijn1990]
[Burke2000] Adapted from Yazicioglu

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Instrumentation Amplifier Overview
(Resistive Gain Element) [Toumazou1989]

Cp I1
x1 x1

IR1 R1 I1 R2

OUT
IN

x1 x1
Cp
I1

Gain Maximal Input Noise PSD (RTI):

Impedance Thermal + 1/f Noise
R2 ν R2
Av = 1 2 ×ν BUF
2
+ν R21 + 2
R1 Av 2
jωC p

Ratio of the Highest possible Noise from Noise from

resistors define input impedance Input Buffers Resistors
the gain Adapted from Yazicioglu

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Instrumentation Amplifiers Overview
(Resistive Gain Element) [Toumazou1989]

OUT
IN

Limited CMRR
Lower-Power Limited
Dissipation DC Headroom
1
σ BUF Unlike in 3-Opamp IA Additional DC blocking
resistor terminals are circuitry might be
Mismatch of the input connected either to required for rejecting
buffers define CMRR input or to the output electrode polarization
voltage
Adapted from Yazicioglu

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Instrumentation Amplifier Overview
(Capacitive Gain Element)
R Sub-threshold
[Harrison2003]: de facto design transistors
[Olsson2003]

C1 C2
–
Cp OTA VOUT
+
C1
C2 R Large Input
Gain
Impedance
C1 1
≈
C2 Noise PSD (RTI): jwC1
Thermal + 1/f Noise
Ratio of the 2 Defined by the
 C1 + C2 + C P  input capacitance,
capacitors define   ×ν OTA
2
C1
the gain
 C1  Adapted from Yazicioglu

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Instrumentation Amplifier Overview
(Capacitive Gain Element)
R

C1 C2
–
Cp OTA VOUT
+
C1
C2 R
Limited CMRR

1
σ C2 + σ C2 + σ Cp
1 2
2
Low-Power Excellent
Dissipation DC Headroom
Mismatch of Inherently AC coupled
passives define Only active component
CMRR is an OTA
Adapted from Yazicioglu

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Operational Transconductance Amplifier
C2

C1 Ma Mb
vin
gm Vout
To reduce 1/f noise, increase the gate area of
vref
C1
input transistors M1 and M2. However, this
Mc CL
C2 increases the op-amp input capacitance Cin. W/L = 0.50
Md

VDD Strong inversion

M7 M8

VcascP
M9 M10 W/L = 200
[Harrison2003] Weak inversion
[Olsson2003] 8 μA Maximum gm/ID vout
Ibias
v- M1 M2 v+
VcascN
4 μA 4 μA

M5 M3 M4 M6

W/L = 0.27
VSS Strong inversion
Greatly reduced gm/ID
Adapted from [Harrison2003]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Measured Neural Amplifier
Input-Referred Noise

SPICE simulation

• Region of “Spikes” Excellent Performance

• Region of EEG/ECG/ECoG/etc, 1/f (process) limited

Recent work expanding on

this core designs:
• [Sarpeshkar2007]
• Good Overview: [Nurmikko2010]

Adapted from [Harrison2003]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Dynamic Compensation: Chopper Modulation
Address Several Amplifier Non-Idealities

[Enz1987]

Chopper Modulator
IA

Modulates
differential signal
Transparent to
Common-Mode
Signals

Adapted from Yazicioglu

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Chopper Modulation
Compensation for Amplifier Non-Idealities

[Enz1987]
Advantages:
Shifts (Modulates)
IA
 1/f noise
 Mismatch related errors
to the out of the signal band
Disadvantages:
• Input Impedance is
reduced (1/jwC)
• Requires DC blocking
circuitry for biopotential
applications

Adapted from Yazicioglu

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Chopper Modulated Instrumentation Amplifiers
(Resistive Gain Element)

• “Simple” addition of switching at front and back end

(Need to consider the bandwidth of micropower amplifier)
• 1/f noise is modulated by the output chopper
• Mismatch errors are modulated by the output chopper
• Requires DC blocking circuitry for biopotential
applications…
Adapted from Yazicioglu

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Chopper Modulated Instrumentation Amplifiers
DC Blocking using a Servo Loop

Adapted from Yazicioglu

[Yazicioglu2006, Denison2007]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Chopper Modulated Instrumentation Amplifiers
DC Blocking using a Servo Loop

Principle Implementation
(Operation)
IA (Resistive Gain)

R1 vin,n
vin,p +
Vin
vin,p
-
IGM

V vout,p
IR = vout,n R2 vout,p +
R vout,n
Vout
Ref I2 I3 I3 I2 -
IV = I R − I F DC Servo

VMAX = I F × R Apply Feedback (AC) as a nulling
input current in front-end

IF IGM

Adapted from Yazicioglu DC Headroom = I GM , MAX × R1

[Yazicioglu2006]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Chopper Modulated Instrumentation Amplifiers
(Resistive: Baseband vs. External Modulation)

+ +
IA
Vin Vout
(Resistive Gain)
- -
Adapted from Yazicioglu

Baseband
TypeAmplifier
I Chopper
TypeArchitecture
III

Gain R2 R1 ＝ R2 R1

CMRR 1 σ BUF < ∞

Noise PSD (RTI) ν thermal
2
+ν 12/ f > ν thermal
2

Input Impedance ≈ 1 jwC p > ≈ 1 j (w ± wchop )C p

Power P = P+Pservo

DC Headroom -- IGM*R1

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Chopper Modulated Instrumentation Amplifiers
(Capacitive Gain Element)

C1 C2
E I – I E
X N OTA N X VOUT
T T T T
+
C1
C2 R

There are several possible architectures for

introducing chopper modulation

Adapted from Yazicioglu

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Chopper Modulated Instrumentation Amplifiers
(Capacitive Gain Element, Internal Modulator)
Adapted from Yazicioglu

To mitigate, off-chip caps

(poor matching)
Beware ofCparasitic
2

C1 high-pass from modulator

[Verma2009]
C1
C2 [Denison2008]

• Chopper modulator operates at the virtual ground

• Same DC blocking scheme as the uncompensated amplifier,
i.e. beyond supply level DC blocking
• Residual offset from the chopper modulator is compensated
by the DC servo
• Amplifier switches at nodes with good dynamic performance
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Chopper Modulated Instrumentation Amplifiers
(Capacitive: Baseband vs. Internal Modulation)
R

C1 C2
+ –
OTA VOUT
- +
C1
C2 R
Off-chip
Adapted from Yazicioglu

Baseband
Type Amplifier
II Int Chopper
TypeArchitecture
IV

Gain C1 C2 ＝ C1 C2

CMRR 1 σ C2 = 1 σ C2

Noise PSD (RTI) ν thermal

2
+ν 12/ f > K 2 ×ν thermal
2

Input Impedance ≈ 1 jwC1 ＝ ≈ 1 jwC1

Power P < P+Pservo+PNoise

＝ 62
DC Headroom Rail-to-Rail Rail-to-Rail

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Chopper Modulated Instrumentation Amplifiers
(Capacitive Gain Element, External Modulator)

Sense DC
of the
[Denison2007] output

Reduced Input Impedance

1
≈
jw(1 ± f chop f sig )C1

Limited DC Headroom Signal Path

 C DC  VDD
  ×
 C1  2
Noise PSD (RTI) (no 1/f)
DC servo path
 C1 + C2 + C DC + C P  2
  ×ν gm
 C1  Adapted from Yazicioglu

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Chopper Modulated Instrumentation
Amplifiers
(Capacitive: Baseband vs. External
R Modulation)
Note work by
[Makinwa2010] to C1 C2
+ –
improve input OTA VOUT
impedance with - +
positive feedback C1
C2 R

Adapted from Yazicioglu

Baseband
Type Amplifier
II Ext Chopper
TypeArchitecture
IV

Gain C1 C2 ＝ C1 C2

CMRR 1 σ C2 < ∞
Noise PSD (RTI) ν thermal
2
+ν 12/ f > ν thermal
2

Input Impedance ≈ 1 jwC1 (

> ≈ 1 jw 1 ± f chop f sig C1 )
Power P < P+Pservo

DC Headroom Rail-to-Rail > (CDC C1 )× (VDD 2)

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 What are the Signals for Detection?
• Cell communication is combination of electrical and chemical activity.
• Electrical Signals  Neural Activity, Muscle, “Derived”
• Chemicals: Neurotransmitters (NT)  learning, memory, mood,
cognition, and neuropathologies (dopamine in Parkinson’s disease
and drug abuse) AND other agents  glucose, etc

~20nm
Scale MATTERS
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Making the Electrode Work Harder for You?
Tissue Conduction is Ionic, Electronics is …

Hypothesized Double Layer @ Charged Exchange at the electrode (Non-polarized)

Electrode-Tissue Interface Electrode-Tissue Interface

What other
reactions are
Hypothesized Structure available at the
Of Double Layer: Polarization Impacts: electrode?
Region “A”: Surface hydration. •Distortion of evoked response.
Region “B”: Loosely held •Energy losses due to polarization
hydration layer with hydrated Cations.
Region “C”: Bulk Solution. impedance.

[Mohseni2009a,2009b,2010]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Emerging Application:Fast-Scan Cyclic Voltammetry
• If the chemical is inherently “electro-active”, can use the recording electrode
• Typical FSCV systems in neuroscience have sensitivity of 10 – 50nA/µM for dopamine.

R. M. Wightman, Chem. Rev., 2008 [Mohseni2009a,2009b,2010]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Time-Share Voltammetry and EPHYS

• Quasi-simultaneous chemical and electrical recording at

the same recording site

• Measuring neurotransmitter and its postsynaptic effect at

the same time and microscopic locale

Triangle Current
Wave (Voltammetry)

Potential
(Electro-
physiology)

Carbon-fiber
Microelectrode

µV
ms
Isolated
Single Units

[Mohseni2009a,2009b,2010]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 What are the Signals for Detection?
• Cell communication is combination of electrical and chemical activity.
• Electrical Signals  Neural Activity, Muscle, “Derived”
• Chemicals: Neurotransmitters (NT)  learning, memory, mood,
cognition, and neuropathologies (dopamine in Parkinson’s disease
and drug abuse) AND other agents  glucose, etc
• “Indirect Measurements” of Patient State?

Scale MATTERS ~20nm

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Sensor Fusion with Accel
Surrogate Markers for “Brain State”?
• Accelerometers: Ideal Sensor for Implant?
 Build off existing experience with pacemaker (rate-responsive
pacing)
 Many biomarkers, challenges can be related to posture, activity
• Groups actively exploring as depression biomarker…
Universal Scaling Law in Human Behavioral
Organization (Phys Rev Letters:99, 2007)

Shut device down

during sleep?

Detect biomarker
for Depression
(poor sleep)

[Denison:Accel2007]
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Implantable Accelerometers Possible?
• Practical 2μW Three Axis Reference Noise, Power FOM
Accelerometer (1μA, 1.9V)
F*rt(W/Hz)
 3-axis Micromachined Sensor
Lemkin[1999] 0.17aF/rtHz, 3.6e-20
 +MEMS design is robust to shock 45mW
(> 20kg), - 1fF/g
 Requires a low noise interface Wu[2004] 0.02aF/rtHz, 3.5e-21
with excellent power 30mW

 Correlated double sampling

Sadat[2005] 0.68aF/rtHz, 1.7e-20
approach
600uW
• Gauging designs on general figure-of-
merits Tavakoli[2003] 2.8zF/rtHz, 4.0e-22
20mW
 Scale designs with equivalent
capacitance sensitivity
Denison[2007] 1.0aF/rtHz, 8.1e-22
 Scale noise appropriately with 650nW
power
 Recent performance mostly from Paavola[2007] 50aF/rtHz, 3.2e-19
large MEMS element 40uW

Kamarainen 70aF/rtHz, 4.9e-20

[2009] 500nW

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Chapter 2 : Summing Up
Sensors and Interfaces

• System design and constraints drive IC design

• Signals are generally “low-bandwidth”
• …but signals are small, prone to confounding inputs
• Multiple sensing domains available, might need to
“fuse” for acceptable specificity
• Detecting signals is “easy,” how do you manage data?

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Typical Presenting Problem: Neural Signals
Require Amplification and Transmission
• Neural signals:
• Action potentials (“spikes”): 100 Hz – 7 kHz, 50-500 μV amplitude
• Local field potentials (LFPs): < 1 Hz – 200 Hz range, up to 5 mV amplitude
• Typical array might have tens to hundreds of signals of interest

Superimposed waveforms of typical

neural action potentials (“spikes”)
recorded extracellularly.

Slide Courtesy of Harrison

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Multi-channel neural recording devices produce lots of data!
Method 1: Use fast MUX and fast analog-to-digital converter (ADC);
transmit exact shape of electrode waveforms
electrode
1
electrode 100 electrodes ×
2 15 kSamples/second ×
electrode 10 bits/Sample
3

time
= 15 Mbit/second

Method 2: Set threshold for spike detection at each electrode;

transmit one bit for electrode every millisecond
1 1
electrode
1
2
electrode
2
100 electrodes ×
electrode
1 kSamples/second ×
3 1 bit/spike
time = 100 kbit/second Reduction in data rate
by a factor of 150
Slide Courtesy of Harrison
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Chapter 3
Algorithms and Data Extraction

Major Themes for Section

• Management of Energy
• Management of Information
• Management of Interfaces

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Traditional Biopotential Signal Analysis Flow
DATA ABSTRACTION

Most JSSC Papers…

CLINICAL IMPACT

Adapted from Yazicioglu

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 TENSION: “OPTIMUM” POINT FOR DIGITIZATION?

[Sarpeshkar1998]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 PRINCIPLES FOR ULTRA-LOW-POWER DESIGN
1. Customize Electronics
2. Adaptive Feedback
Architectures
3. Analog preprocessing before
digitization
4. Slow-and-Parallel
5. Subthreshold operation
6. Balance Computation and
Communication Costs
7. Combine inspiration from
biology with perspiration from
engineering

Robustness and Flexibility always trade off

with efficiency. You must have as many extra
degrees of freedom needed to maintain both but
not unnecessarily many.
Analog and Digital should work symbiotically
with digital providing flexibility and robustness
and analog providing efficiency.

Slide Courtesy of [Sarpeshkar1998]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 SNR
45dB SNR
50dB

SNR
40dB

Must Also Consider Sense-Actuation Interactions (>120dB!)

79

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Consider a Typical Use-Case: Cardiac Filtering
• Match frequency content of electrogram to the
bandpass of the filter to increase likelihood of proper
sensing…
• Very similar requirements for “brain sensing”, EMG, etc

Extract this complex

Confounding information

T-wave “oversensing”
From: Ellenbogen, Kay, Wilkoff.
Clinical Cardiac Pacing and Defibrillation.
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Analog Signal Processing
“Power Efficient” Frequency Analysis
(f+Δ)(0o)
LPF
(BW/2)

Amp X2

In-phase LPF

EEG
f(0o)

Quadrature

Amp X2

LPF
(BW/2) ∞ 2
1
 x(t )cos(2πft )dt
o
(f+Δ)(90 ) I=
Energy Spectral Density 2π −∞
∞ 2
1 1 +
φ( f ) = X(f ) =  ( )
2 −i 2πft
x t e dt
2π 2π −∞ ∞ 2
1
Q=
2π  x(t )sin (2πft )dt
−∞

[Denison2008]
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Chopper + Heterodyning Approach
(f+Δ)(0o)
LPF
(BW/2)

Amp X2

In-phase LPF

EEG f(0o)

Quadrature

Amp X2

LPF
(BW/2)
(f+Δ)(90o)

Harmonic
Frequency
f f
of interest
[Denison2008]
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Improved Analog Frequency Analysis
Example: Application to Cardiac Signals

<1.25 μA

Calculate Signal Power

Single Instrumentation
after ADC to sustain
Amplifier for both
dynamic range
channels

[Yazicioglu2010] (2 )
N −bit 2
= 22 N
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Power Efficient Frequency Analysis
Example: Application to Cardiac Signals

Threshold Band-Power

ECG

Analog Commercial Microcontroller (or Custom DSP)

[Yazicioglu2010] 84

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Now Consider the Counter-Point!

Implementation of Digital-Based
Algorithm Methods. DSP toolkit rapidly expanding…
Operation Details Application

128-tap Heart rate detection

FIR
23-tap Pulse-oximetry

1024-point Heart sound processing

FFT/IFFT 512-point Speech processing for cochlear

implant

sqrt(x) Heart rate detection

atan(x), sin(x), Heart sound processing

cos(x)

Find max. Heart rate variability

Sort data measurement
Median filter Heart rate detection

Slide Courtesy of [Kwong2010]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Signal Processing Platform Overview
1. Accelerators for efficient biomedical signal processing
2. Functions from 1V to 0.5V, Process node is 130nm
3. Power gating to reduce leakage of idle components
SW Debug
Support

JTAG

Slide Courtesy of [Kwong2010]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Application 1: EEG Feature Extraction [1]

Note: Still need to implement classifier

Energy
VDD
(per 512 inputs)
CPU 1.0 198µJ
Accel. 0.7 19.3µJ

Reference [Shoeb2004] & [Denison2009b] Reduction 10.2x

for discussion of the algorithm goals
and classification implementation
Slide Courtesy of [Kwong2010]
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Application 2: EKG Feature Extraction [2]

QRS complex

Energy
VDD
(per beat)
CPU 1.0 188µJ
Accel. 0.7 16.4µJ
Reduction 11.5x

Slide Courtesy of [Kwong2010]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Having a Run-Off? Validating the Signal Chain
• Regulatory requirements say we are not done yet…
• Verification: Did we build what we said we would build?
(Easy to execute: Checking specs like noise, power, etc)
• Validation: Did we build what actually needs to be built?
(This can be much harder to quantify and prove)
• Resources for Annotated Data  Benchmark your System!
 Physiobank: signals of interest: EEG/ECG/EMG/BMI…
http://www.physionet.org/physiobank/
 For EEG there are currently two intracranial EEG databases and one scalp
EEG database (Ali Shoeb, MIT/MGH):
• The flint hills intracranial EEG database:
http://www.fhs.lawrence.ks.us/PublicECoG.htm
• The Freiburg intracranial EEG database: http://epilepsy.uni-
freiburg.de/freiburg-seizure-prediction-project/eeg-database
• The MIT-CHB scalp EEG database:
http://www.physionet.org/pn6/chbmit/
 TEST on ALL the DATA

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Example: Preclinical Validation
of “Brain State” Detection

Adapted from
[Denison2009a,2011]

Research Hybrid LFP Processor

Research Device (Sense + Alg + Stim)

Validation Protocol & Methodology

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Typical Sample Run
Real-Time Cursor Control

Adapted from
[Denison2011]

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Adapted from
Quantifying Detection Performance
[Denison2011a] Algorithmic Classifier Boundary
PDF
Brain State 1 Brain State 2

1.1μVrms
mean separation
for cursor control

Physical Attribute “Marker” (e.g. µVRMS, Inband Power)

Trade-off Point

10μW/channel
10 year implant 0.5

0
Comparing Receiver Operating Characteristic 0 0.5 1
False Positive Probability
(See reference [ROC-Primer] for Overview)
© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE
 Chapter 3 : Summing Up
Algorithms and Classifiers

• Management of Energy  Information  Interface

• System design and constraints drive IC design
Analog vs. Digital Signal Processing vs. Telemetry
• Verification of circuits is only a small part of the effort…
We must also VALIDATE that our solutions work

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Chapter 4: Putting it Together!
State of the Art Systems Examples

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Status of Actuation Methods
in the Nervous System (2011)
OCD1 Depression2
775,000

IN DEVELOPMENT
Parkinson's Disease
216,000 Epilepsy2
245,000
COMMERCIAL

Essential Tremor
80,000 Neurodegenerative
Diseases (drug-
Dystonia1 device)3
3,500,000
Severe Spasticity
1,200,000 Migraine
Headache Pain3
Chronic Pain 904,000
1,300,000
Nonopioid
Gastroparesis1 Chronic Pain4
653,000
Overactive Bladder
and etention Fecal Incontinence2
1,800,000 692,000

1. Humanitarian Device Exemption (HDE), 2. Investigational Use Only (IDE), 3. Research,

4. Investigational New Drug Patient #’s = US Net Prevalence (indicated, addressable population)

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Spinal Chord Stimulation: Unmet Need

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 New System Architecture
Responsive Neurostimulation

Control Algorithm &

Feedback “Actuation”
!
Potential
Contamination
Power Source

Biomarker Sensor
With
Primary
Signal System
Disturban-
ces & Sensing IC Interface Processing Control
Noise Interface

Calibration Data Data External

Signal? Storage Transmission Unit

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Technical Solution for Unmet Need
“Responsive System” (CE-Marked, Investigational Use Only US)

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Our Other Mandate:
Enabling the Next Generation of Scientific Discovery

Using silicon circuits to “debug” and potentially help address disease

in neural circuits

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE

 Closing Summary Thoughts…
• Effective interfaces to biological systems requires a holistic approach…
Actuation of Biological Systems
• Always a strong bias for addressing issues with compensatory technology
• Actuation offers many of the most exciting opportunities for technology: scaling of
systems, sophistication of methods, and fundamentally new techniques
Sensors and Interfaces
• One of the key components in biomedical systems is the analog front-end (defines
signal quality)
• Driver for minimal power budget has lead to several new topologies during recent
years…
• And many novel schemes for sensing activity (Accelerometers, biochemical, etc)
Algorithms and Classification
• Smarter systems require more signal processing which can often dominate the
system power
• Circuit designers need to get comfortable with validating complete designs to
have more clinical impact
Total Systems Solutions
Putting all the pieces together results in the highest impact, both for near-term
applications as well as instruments for scientific discovery

© 2011 IEEE IEEE International Solid-State Circuits Conference © 2011 IEEE