Neuromorphing - Building Brains in Silicon (BE 526)  

Kwabena Boahen, PhD
Department of Bioengineering
boahen@seas.upenn.edu
 


Description: We introduce neurobiologists to the physical constraints on neural computation—namely, noise, wiring, and energy. We introduce engineers to the unrivaled performance of biological systems—achieved by using physical resources efficiently. We pursue these goals by studying large-scale models of entire neural systems, consisting of thousands of silicon neurons that respond in real-time. Students conduct analytical (deriving mathematical solutions), computational (simulating circuit behavior), and experimental (testing prefabricated chips) exercises. They work in multidisciplinary teams, combining their expertise in neuroscience and neuroengineering, and thus rudimentary knowledge of one or the other is fine.

Prerequisites: Students with advanced knowledge in neurobiology but rudimentary knowledge in electrical engineering or vice versa are welcome. Biology students should have a course in Cellular Neurobiology (e.g., BIOL251 or INSC572). A course in Systems Neuroscience (e.g., BIOL451 or INSC573) is recommended but not required. Engineering students should have a course in Solid-State Circuits (e.g., EE319). A course in Integrated Circuits (e.g., EE419, 560, or 562) is recommended but not required.

Goals: To capture the structure and function of entire neural systems in real-time using microelectronic devices. To build these neuromorphic models, we proceed from the device level, through the circuit level, to the system level. At the device level, we draw parallels between electrodiffusion of electrons through transistor channels and electrodiffusion of ions through membrane channels. At the circuit level, we implement synaptic interaction, dendritic integration and active membrane behavior using transistors. At the system level, we synthesize the spatiotemporal dynamics of the cochlea, the retina, and networks of spiking neurons in cortex.

Textbooks: None required. But, the monograph, Analog VLSI and Neural Systems, by Carver Mead, is a good introduction to Very Large Scale Integrated electronic systems. The book, From Neuron to Brain, by Kuffler, Nicholls and Martin, is a good introduction to the brain.

Grading: Based on individual homework and team lab reports (groups of two).

Target Audience: This course is intended to draw advanced students from multiple disciplines with an interest in bridging disciplines. Students are encouraged to pool their expertise in different areas in teams of two.

Topics:
Overview
    VLSI CMOS Technology: From Transistors to Multichip Systems
    Systems Neuroscience: From Ion-Channels to Perception
Electrodiffusion
    Ion-Channels: Electrodiffusion in Liquids
    Transistors: Electrodiffusion in Solids
    Parallels between Ion-Channels and Transistors
Synaptic Interaction
    Single-Transistor Gap Junctions
    Single-Transistor Chemical Synapses
Temporal and Spatial Integration
    Dendrite/Soma Model: Diode-Capacitor Dynamics
    Cell-Syncytium Model: Laplace’s Equation
Active Membrane Properties
    Spike-Generation: Fast Na- and K-Channels
    Spike-Rate Adaptation: Ca- and Voltage-Dependent K-Channels
    Burst-Generation: T-type Ca-channels
Spatiotemporal Dynamics
    Finite-Element Analog of Basilar Membrane and Cochlear Fluid
    Anatomically Detailed Models of Outer and Inner Retina
Single-Chip Systems
    A Silicon Cochlea: Sharp Tuning and Gain Control
    A Silicon Retina: Luminance Adaptation and Contrast-Gain Control
Multichip Systems
    A Silicon Optic Nerve: Interchip Communication using Address-Events
    A Silicon Visual Cortex: Orientation Hypercolumns

Files:
   Course Calendar with Links to Reading Assignments
Linked BE526_Calendar05
   Course Handouts
BE526_Outline05.pdf Course Outline
BE526_Calendar05.pdf Course Calendar (no links)
Tutorial.nb Mathematica tutorial
BE526_Design_Course05.pdf Design course description
   Homework
hw1.pdf Home Work 1
hw2.pdf Home Work 2
hw3.pdf Home Work 3
RetinaAnalysis.nb Mathematica simulation for outer plexiform layer model
RetinaHW.nb Mathematica homework for outer plexiform layer model
CochlearSimulation.nb Mathematica simulation for 2D cochlea model
CochlearHW.nb Mathematica homework for 2D cochlea model
   Lab Assignments
Lab1.pdf Introductory Lab
Lab2.pdf Gate-Source Dependence
Lab3.pdf Early Voltage and Transistor Gain
Lab4.pdf Modeling Synaptic Currents
Lab5.pdf A Simple Neuron
Lab6.pdf Spike Timing: An Auditory Neuron Specialization
Lab7.pdf Current-Divider and Diffusor
Lab8.pdf Silicon-Neuron Array
Lab9.pdf Synchrony in Inhibitory Recurrent Networks
Lab10.pdf Silicon Growth Cones (requires notebook)
   Lab Supplements
1n4148.pdf Diode Specifications
2400.pdf Meter/Supply Specifications
6517.pdf High Impedance Meter Specifications
n9cn-params.txt MOSIS Parametric Test Results: chip lot N9CN
 
Last modified 18, January 2003