Understanding the brain is one of the major scientific goals of the 21th century. In the bottom-up approach, this means looking at biological neural networks and trying to figure out their principle of operation. Neuroscientists all over the world are recording the membrane potentials of neurons, in vitro as well as in vivo; still it is not possible to listen to more than a few neurons at once. This has led to the situation that numerical models, solved on digital computers, have become the main method in testing new theories of the behavior of neural microciruits. Analog Neural Networks implemented in VLSI technology are the only feasible alternative today.
Modeling neural microcircuits in analog VLSI - this is the intend of the current chip development in the Heidelberg Electronic Vision(s) group. Contrary to the HAGEN chip, these spiking neural network ASICs mimic neural behavior to a much larger extend.
Our first hardware model close to the biological example is the Spikey chip developed in the framework of the Sensemaker project. It will mimic the behavior of small neural circuits, including some aspects of synaptic plasticity.
The waferscale integration apporach will increase the complexity of the VLSI models by several orders of magnitude. The goal are networks with billions of synapses operating 10000 times faster than biological nervous systems.