Mark A. Glover

Analogue VLSI Leaky Integrate-and-Fire Neurons and Their Use in a Sound Analysis System

Integrate-and-fire neurons are simple model neurons which can handle continuously time-varying signals. We have applied them to problems in real-time analysis of sounds. Two different chips have been built: the first had a fixed network architecture with all synaptic weights identical, and the second is reconfigurable with individually programmable weights. We present results characterising the latter chip, and results from processing real data from the earlier chip. We note that the second chip provides a more general integrate-and-fire neuron implementation.

Analogue VLSI for temporal frequency analysis of visual data

This paper introduces an algorithm capable of analysing temporal frequencies present in any visual scene. By estimating the frequency of temporal variations, it is possible to distinguish between the objects that create them. Simulation results highlighting the operation and performance of the algorithm are presented, along with results from a test chip. The chip was fabricated in a 0.6 /spl mu/m CMOS process and implements the first phase of the complete algorithm.

SPIKEII: an integrate-and-fire AVLSI chip.

We present the SPIKEII chip, an integrate-and-fire neural network chip with programmable synapses implemented in analogue VLSI. It is the successor to the SPIKEI chip. We describe the circuitry, and show some results using networks of integrate-and-fire neurons.

A comparison of a hardware and a software integrate and fire neural network for clustering onsets in cochlear filtered sound

Onset clustering (which we use as part of a system for sound segmentation) uses integrate-and-fire neurons to perform across spectrum and across time clustering of increases in sound intensity in different parts of the spectrum. We show that a network of recently developed analogue VLSI integrate-and-fire neurons can perform this task in real-time, and compare its performance with a simulated network.

Minipix: focal plane temporal frequency image processor

Previous research has highlighted the potential utility in analysing temporal frequency signatures. This paper introduces an improved algorithm capable of accurately encoding the fundamental frequency of such temporal signatures in the form of a robust pulse train, while consuming minimal power and area. The approach makes use of focal-plane computation techniques to produce a compact pixel-processor, consuming 60 /spl mu/m by 60 /spl mu/m when implemented in a 0.6 /spl mu/m process, with a fill factor of 14.1%. The adopted analogue circuit techniques are biased in the subthreshold region of operation, giving a power consumption of less than 100 nW per pixel when operating at 1 kHz from a 5 V power supply.

Using analogue VLSI leaky integrate-and-fire neurons in a sound analysis system

Integrate-and-fire neurons are simple model neurons which can handle continuously time-varying signals. We have applied them to problems in real-time analysis of sounds. Two different chips have been built: the first had a fixed network architecture with all synaptic weights identical, and the second is reconfigurable. We present results characterising the latter chip, and results from processing real data from the earlier chip. We note that the second chip provides a more general integrate-and-fire neuron implementation.