Dendrites, Difusors, and WTA
Overview:
- One key objective in this lab is looking at the behavior and modeling
of dendritic components and how they come together for computation.
-
One objective is to
explore the behavior of an elegant current-mode implementation of a
resistive network called a diffusor network and understand
some of the things that these circuits are used for.
Resistive networks are often used in biological systems for a variety
of tasks. In particular, they can be used to perform
local spatial averaging .
Such averaging is used to improve the signal-to-noise
ratio or to obtain a local reference to which signals can be
compared.
In this lab we will examine a one-dimensional resistive network made
up of diffusors.
The diffusor relies on the exact
linear relationship between channel current and the carrier concentration
in subthreshold conduction.
Reading:
- Key lab material available ( pdf ).
- First dendritic transistor channel modeling paper,
including reconfigurable capability as part of the resulting architecture
( pdf ).
-
Dendrite Channel Model with direct comparisons / extensions to Rall's formulation
( pdf ).
All of the data was compiled on an FPAA device.
Note the use of amplifiers (FG OTA of gain between 20-30) for the data.
-
Discussion of Dendritic Wordspotting related to HMM classifiers
typical for speech recognition
( pdf ).
All of the data was compiled on an FPAA device.
Note the use of amplifiers (FG OTA of gain between 20-30) for the data.
- Board Pictures
Project Questions:
The focus of this project is to get a working
dendritic line,
built from a diffusor circuit approach
with biasing approximating a dendrite cable diameter changes,
and then pulling multiple lines together to visualize characterization behavior.
Adding Synapses and Dendrite Diameter
This section discusses adding of synapses and dendritic diameter to the initial model.
We started in a previous case that we utilize additional pFET switch devices,
which are potentially already compiled and used as current sources in the previous case,
where you put in a waveform in the pFET source to emulate the synaptic concepts.
Second, you will shift the effective diameter down the line,
from distal to proximal direction.
Remember, this effectively shifts the horizontal conductance to increase
at each pFET horizontal tap,
effectively giving a linear decrease in the floating-gate voltage at each tap.
Bulding a Dendritic-Modelled Neuron Classifiers
This section discusses duplicating one dendrite into two dendrites plus an additional compartment
to illustrate a first dendritic-modeled neuron classifier.
You will start by building the two lines,
one you have already used,
an additional tap,
to create the classifier structure discussed by the Dendritic Wordspotting structure above.
You will want to generate a sequence of inputs to demonstrate the "yes" and "no" classification.