Transcript Feature-based Attentional Modulation Depends on Task

Analog VLSI Neural Circuits
CS599 – computational
architectures in biological vision
Charge-Coupled Devices



Uniform array of sensors
Very little on-board processing
Very inexpensive
CMOS devices



More onboard processing
Even cheaper!
Example: ICM532B from www.icmedia.com: single-chip solution
includes photoreceptor array,
various gain control and color
adjustment mechanisms, image
compression and USB interface.
Just add a lens and provide power!
The challenge

Digital processing is power hungry

Analog processing is much more energy efficient

But … so much variability in the gain of transistors
obtained when fabricating highly integrated (VLSI)
chips that analog computations seem impossible:
nearly each analog amplifier on the chip should be
associated with control pins, analog memories, etc
to correct for fabrication variability.
Hopeless situation?
A VLSI MOS transistor
An analog chip layout: the wish
An actual chip: the cold reality
Biological motivation



Well, there is also a lot of variability in size and
shape of neurons from a same class
But the brain still manages to produce somewhat
accurate computations
What’s the trick? online adaptability to counteract
morphological and electrical mismatches among
elementary components.
Remember? Electron Micrograph of a Real Neuron
Mahowald & Mead’s Silicon Retina
Smoothing network: allows system to adapt
to various light levels.

Andreou and Boahen's silicon retina
See http://www.iee.et.tu-dresden.de/iee/eb/
analog/papers/mirror/visionchips/vision_chips/
andreou_retina.html
Diffusive network



dQn/dt is the current supplied by the network to node n,
and D is the diffusion constant of the network, which
depends on the transistor parameters, and the voltage Vc.
Full network


Two layers of the diffusive network: upper corresponds to
horizontal cells in retina and lower to cones. Horizontal Nchannel transistors model chemical synapses.
The function of the network can be approximated by the
biharmonic equation
where g and h are proportional to the diffusivity of the
upper and lower smoothing layers, respectively.
Full network
VLSI sensor with retinal organization
Carver Mead: the floating gate
www.cs.washington.edu/homes/hsud/fg_workshop.html
Spatial layout
Electron tunneling
Electron tunneling
Hot electron injection
Hot electron injection
Spatial layout
A learning synapse circuit