From single-molecule physics to

neural interfaces the case for

CMOS bioelectronics
Ken Shepard
Bioelectronic Systems Laboratory
Columbia University, New York, NY

Interfacing the solid-state and the

biological
99% of time light is used as the intermediary, both for in
vivo imaging and molecular diagnostics

Advantages
Allows non-invasive interrogation.
Diversity of chromophores (organic
and inorganic) have been developed
for sensing
Optogenetics has allowed for direct
actuation

Disadvantages
Very shot-noise-limited channel. Most
pronounced at high bandwidth
Weak signal levels from single optical
reporter.
Indirect measure of biophysical
processes
Photobleaching limits measurement
time
Complex instrumentation

Interfacing the solid-state and the

biological
Direct electrical transduction requires close proximity

Form factor mismatch

Biological is round and squishy; solidstate is hard and flat.
These can be reconciled in two ways:
Miniaturize one with respect to
the other.
If of the same size scale, make
one conform to the other.

Signaling system mismatch

Biological systems use biochemical and
ionic signaling; solid-state systems use
electrons
These can be reconciled in two ways:
Sense molecules through their
charge or redox properties
Convert ions to electrons with
electrochemical reaction
Require gain in electronics, transducer,
or both.

Interfacing the solid-state and the

biological
Advantages of CMOS integration.
Reduction in parasitic capacitances
Signals are usual transduced to small
(< 10 nA) currents.
Keeping capacitances low is essential
to achieving high noise-limited
bandwidths.

Enables dense arrays

Recover the functionality of CMOS
imagers
This can be used for imaging (spatial
extent on array corresponds to spatial
extent in sample) or multiplexed
diversity
Time-multiplexing of amplifiers
required to achieve density.

Interfacing the solid-state with the

biological
Single-molecule bioelectronics
Nanopores and single-molecule field-effect
transistors
CMOS integration into array

Electrochemical imagers
Hybrid ion-channel-CMOS systems
Neural interfaces

Interfacing the solid-state with the

biological
Single-molecule bioelectronics
Nanopores and single-molecule field-effect
transistors
CMOS integration into arrays

Electrochemical imagers
Hybrid ion-channel-CMOS systems
Neural interfaces

Single-molecule versus ensemble

Continuous quantities become discrete; static quantities can become
dynamic
Microscopic mechanism not observable with ensemble become evident.

Ensemble DNA Melting Curves

Single-molecule hairpin FRET

(Ririe et al, Analytical Biochemistry 1997)

(Grunwell et al, JACS 2001)

Signal levels for single-molecule

fluorescence
Photons are discrete; weak
optical signals exhibit shot
noise
Typical fluorophores emit
2500 photons/sec = 0.4 fA
of current
Resulting Nyquist
bandwidth with this signal:
~ 1.25 kHz at 1-bit
resolution
~150 Hz at 4-bit resolution
~ 10 Hz at 8-bit resolution

Single-molecule molecular diagnostics

Most current commercial genomic molecular diagnostics are
ensemble-based and fluorescence-based (qPCR, Illumina HiSeq)
Huge impetus to push genomic molecular diagnostics to singlemolecule
Eliminate the need for a template amplification phase prior to
sequencing, which can introduce bias and substitutions
Significantly reduces sample preparation complexity.

Fluorescent-based detection is very signal-challenged when

scaled to single-molecule.
Pacific Biosciences Single-Molecule Real-Time (SMRT)
sequencing systems

Single-molecule transduction
Good bioelectronic single-molecule transducer
has to have two properties: high gain and spatial
localization of this gain.
Two transducers have this property: nanopores
and single-molecule field-effect transistors.

Biological nanopores
Signal levels ~ 10 pA

Solid-state nanopores
Signal levels ~ 1 nA

1pA = 6,200,000 e-/sec

smFETs
Signal levels ~ 10 nA

Signal levels lost to lousy electronics

Input capacitance
> 15 pF
Bandwidth limited
to 100 kHz
Set up for
electrophysiology

Fast translocation rates

20 Mbases/sec typical for
solid-state pores (slightly
slower for biological pores)
Standard approach today is
to put an enzyme in front to
slow down translocation
rate.
Lets adopt a different
approach: lets speed up
the electronics!

Lower Noise Can Afford Wider Bandwidth

amplifier feedback

amplifier input

CF

wiring/fluidics
electrolyte
charge distribution

nanopore
& membrane

RF
RA
CM

RP

filters to flatten
freq response
& antialiasing

CW CI

Filters

vn
VBIAS

Voltage-clamp amplifier

J. K. Rosenstein, et al, Nature Methods 9:487 (2012)

Multi-channel CMOS Preamplifier

Aluminum

100 MW

Electroplate
Silver

Etch
Aluminum

Chlorinate
Silver (Ag/AgCl)

High-Conductance, Low-Capacitance Nanopores

Rosenstein et al, Nature Methods 2012

Baseline / Open-headstage Noise

Add the pore

Membrane capacitance
Flicker noise of pore

Fast DNA Translocation Events

Events
shorter
than 10
msec clearly
attenuated

3.5nm pore, 600mV, 25bp dsDNA

Intra-Event Structure

Squeezing out the required temporal

resolution

Approaching 20 MHz..

smFET sensor

Essential attributes of a single-molecule

transducer
Spatial locality (~ defect size)
High transducer gain (field-effect
modulation of transmission)
S. Sorgenfrei, et al, Nature Nanotechnology
Specificity (covalent attachment of probes)
6:126 (2011)

Thermodynamics from time series data on single

molecule
DNA 1:
Ratio of Area under
Gaussian Fits

Melted
State

Hybridized
State
DNA 2:

Thermodynamics

Multistate time-series data analysis

H

CNT

States
H: hybridized
M: melted
CNT: hybridized interaction with
surface
CNT and H states indistinguishable with
same conductance

Debye screening
Fluctuation amplitude for different target DNA
Remove closest base of target DNA
Experiment at 17C (away from melting point)

smFET arrays
Single-molecule assay

Single-molecule sequencing
Oxford MinION

CMOS smFET arrays have the potential to do for genomic assays what nanopores
are trying to do for sequencing.

Interfacing the solid-state with the

biological
Single-molecule bioelectronics
Nanopores and single-molecule field-effect
transistors
CMOS integration into array

Electrochemical imagers
Hybrid ion-channel-CMOS systems
Neural interfaces

Electrochemical imaging array

No direct optical imaging
approach available.

Pseudomonas aeriginosa

D.L. Bellin, et al, Nature

Communications 5:3256 (2014)

Pseudomonas aeruginosa PA14 is an

opportunistic pathogen that produces
redox-active metabolites called
phenazines which have drastic effects
on community behavior in colony
biofilms. Clinical relevance in
understanding biofilms.

Electrochemical Characterization
of Phenazines

Mutant strains allow individual phenazines to be electrochemically characterized.

Electrochemical imaging of biofilms

Electrochemical imaging of PYO concentration.
Scale Bar = 5 mm

Spatial resolution
limited by lateral
diffusion.

Second generation imager chip

Working
Electrode
Array

Full 1824-electrode
array.
Removal of agar film
will improve spatial
resolution.
Frame rates in excess
of 1 frame/sec
possible.
Scale Bar = 1 mm

Membrane

Working
Electrode
Array

Colony

IC

Interfacing the solid-state with the

biological
Single-molecule bioelectronics
Nanopores and single-molecule field-effect
transistors
CMOS integration into array

Electrochemical imagers
Hybrid ion-channel-CMOS systems
Neural interfaces

Biomimetic systems
Lots of funding in the area of neuromorphic computing.
DARPA Synapse and European Union Human Brain Project
40-year-old efforts in biomimetic systems
Very difficult because fundamentally the technologies (solidstate and biological) are very different.

Whats happening in the synthetic biology world?

Biobricks attempt to emulate techniques used in the design of solid-state
systems.
Artificial genome cell as state-of-the-art
Input/output nearly always fluorescence

Cell with artificial genome (Gibson, et al, Science, 2008)

Fundamentally different capabilities

Biological systems are incapable of certain engineered functions.
Solid-state systems are incapable of certain engineered functions.

Try to get a biological system

to do this.

Try to get a solid-state system

to do this.

Capabilities of solid-state systems

Senses: sight, touch, smell, taste
Energy: solar, electrochemical, biochemical,
mechanical
Solid-state systems can convert photons to electrons
and through piezoelectric materials, convert
mechanical energy to electrical energy.

Ion channels as transistors

Ion channels are transistor-like in
functionality and create the natural
means
to interface the biological with the solidstate.
CMOS electronic interfaces allow us to
exploit these channels in new ways.

Interfacing to single ion channels

Fastest ion channel
recordings ever
made.

1 MHz bandwidth -> 10 MHz bandwidth

Fast opening and closing dynamics of

Alamethicin.

Ion channels act as transistors and produce gain.

Can also be used to study transporters and be formed into arrays.

J. K. Rosenstein, et al, Nano Letters 13: 2682 (2013)

Interfacing to single ion channels

Mass spec of PEG in a-HL

Examining a collection of different length PEG

molecules through a biological nanopore
on a CMOS chip.

First system powered by isolated ion

pumps

pW-scale system

Ion pump characterization

Before the addition of
ATP
Load line

After the addition of

ATP

(Rosemann, et al, in review)

Interfacing the solid-state with the

biological
Single-molecule bioelectronics
Nanopores and single-molecule field-effect
transistors
CMOS integration into array

Electrochemical imagers
Hybrid ion-channel-CMOS systems
Neural interfaces
The meso
problem

CMOS electrophysiology
Goal of integration is significant scaling of
electrophysiology functionality
Multiplexing in manner similar to CMOS imaging

In vitro
NeuroChip

In vivo
NeuroProbe

Neurochip
Goal is study of the retina which conforms to a planar configures
Specifications:

65,536 channels
25.4-mm electrode pitch
14-mm electrodes
Record and stimulate
8 mV rms noise to 15 kHz

256 x 256
front-end

8.4 mm

Science goals
We are looking at the inner retina with the most
complicated wiring, about which almost nothing is known.

Science goals
In particular, there are amacrine cells with very long range
dendrites. These cells modulate the responses of 100s of retinal
ganglion cells.
To understand what these amacrine cells do, we would like to
formulate an input-output transfer function for them.

Science goals
Experimenter-defined light stimuli (inputs) + complete
readout off all effector neurons below these amacrine
cells (outputs) so we can build the transfer function.

Post-processing
1. IBM stock

2. Masking

3. Passivation etch

4. Remove mask 5. Oxide removal 6. HfO2 deposition

Passivation of electrodes
Capacitive interface

Instrumentation
Data rate almost 3 GB/s.
Processed chip

Packaged chip (MEA)

FPGA #4

FPGA #3

MEA in socket
FPGA #2
Circuit board

FPGA #1
18,000 lines of Verilog; 38,000 lines of C++

Neural probe chip: electronics for in vivo

shank electrodes

Caltech shanks

1024 channels
125-mm electrode pitch
Multi-function: voltage
recording, stimulate,
current recording,
current/voltage course
Low-power
8 mV rms noise

6 mm

CMOS can be rendered flexible with

extreme thinning
Up to 2% strain
can be tolerated in
a fabricated CMOS
integrated circuit.
With aggressive
thinning to < 7 mm,
tolerable radii of
curvature can be
as little as 200 mm.

Thinned 45SOI Chip

Chips are like tissue paper and optically transparent. These plots
are from the backside.

51

mECoG (micro-electrocorticography)
with flexible CMOS
256 x 256
front-end

8.4 mm

(Viventi, et al, Nature Neuroscience, 2008)

Conclusions
Go after photons where they are weak: single-molecule, high
bandwidth, direct measure of biophysical processes
Eliminating photons in the interface between biological and solidstate systems implies a number of challenges.
Form factor mismatch must be handled by miniaturization or
through conformation.
Electrical interfaces either detect molecules through their charge or
electrochemical activity or sense ions through an electrochemical
conversion to electrons.
CMOS integration provides for low-noise measurements and
multiplexing.
Highlighted many examples: nanopores, smFETs, electrochemical
imagers, hybrid CMOS-ion-channel systems, neural interfaces.

Acknowledgements

Collaborators:
Marija Drndic, University of Pennsylvania
Jon Viventi, NYU
Rafael Yuste, Columbia
Michael Roukes, Caltech
Andreas Tolias, Baylor College of Medicine
Thanos Siapos, Caltech

NIH R01-HG006882
NIH R01-HG006879
NIH R01-GM107417
NIH U19-AI109761
NIH U01-NS090596
NIH R43-HG007871
Keck Foundation
ONR
N000141310375

Lars Dietrich, Columbia

Virginia Cornish, Columbia
Ruben Gonzalez, Columbia
Colin Nuckolls, Columbia