Peter van der Made
BrainChip Inc., USA
Introducing the Akida Spiking Neural Network Reconfigurable NeuroSoC
Abstract
The Akida architecture is a progression of the SNAP64 technology that we developed 4 years ago. The SNAP64 was a test chip, containing 4 blocks of 64 neurons with 256 synapses each – a total of 65536 synapses. The BrainChip Akida chip and development board are offering a flexible architecture, a Neuromorphic System on a Chip that contains many components that are useful in practical applications, such as convolution blocks, max pooling, programmable spike generators (Gaussian Receptive Fields) for data conversion, local ARM processor, several interface busses, and video frame to spike converters in addition to a large ‘fabric’ of convolutional and fully connected, autonomously learning spiking neurons.
Biography
Peter has been working on Neuromorphic systems since 2004. In 2008 he filed a patent describing a digital Spiking Neuron processor that was subsequently granted. Peter started his career in 1974, after completing his education at the Technical University of Amsterdam. Because of his interest in digital electronics he started work at Sperry Univac in Sydney, Australia. Sperry Univac sponsored him to attain qualifications in digital logic and programming. Sperry Univac was in 1974 the 2nd largest manufacturer of main-frame computer equipment. Because of his interest in the emerging micro-computer field, he joined Olivetti when an opportunity presented itself to dive deeper in microprocessors. Over the next few years he worked at several major companies and he became skilled in hardware and chip design. Pushing the boundaries of technology, he designed a graphics accelerator chip in the early 1980’s. This graphics accelerator chip had a resolution of 900 x 1200 pixels in 4096 colors and used a 20 inch screen. The graphics processing core was clocked at up to 120 MHz which was very high at that time. From 1988 to 1994 he worked at a university teaching computer science and digital design. After 1994, Peter turned his attention to the growing problem of computer malware with the development of a new detection method based on real time behavior analysis and classification. This work resulted in three granted patents. By the end of the millennium he had created a working prototype of the ‘Analytical Behavior Method’ to detect known and unknown malware. By 2002 the company was acquired by Internet Security Systems (now IBM) in Atlanta, with great benefited to the shareholders. BrainChip was founded in 2004 with the aim to develop a better method for machine recognition based on biology. Much time was devoted to the synapses, these tiny spots that occur on the body of neural cells in the brain. These tiny structures hold the key to learning and brain organization. This development was driven by the fact that the brain is a huge spiking neural network and the ultimate learning and recognition engine. The devices that were developed perform autonomous learning and feature extraction. Akida is the cumulation of this work. The research is documented in a book, published in 2012, called “Higher Intelligence”.