Category: Training

IBM has revealed two new artifical brain chips “evolved” from ways that a rat thinks, a cat reacts and a human is wired. Press release

Reverse engineering the brain using nanoelectronic circuitry has the potential for a million times faster processing than a normal brain

but also a million times better energy efficiency than a normal CPU.

The reason we have not engineered our computing systems this way before is that we have not been able too analyze, understand and model synaptic systems in sufficient detail until only a few years ago. For a good introduction to the reverse engineering of synaptic systems, see the excellent talk by Dharmendra Modha, manager of cognitive computing at the IBM Almaden Research Center.

By switching from traditional binary-logic to these massive-synaptic systems, the design criteria for nanoelectronic circuitry changes dramatically: Instead of requiring that every single cell in the IC fabric behaves exactly the same, it is enough that it is possible to have some paths through the fabric that can generate sufficient signal strength compared to the background noise from all other paths. Any design that generate a stable nonlinear response from the selective summation of many input signals will work, as long as each of its inputs includes a memory of the response it generated when it was last active. This means these nanosynapses can be made much smaller than traditional nanologic circuits, and that manufacturing processes can be much more simple and innacurate. In fact it will be more robust to have manufactoring processes and/or circuit designs that are generating a fairly large variation because that makes the system more adaptive!

Quflow has researched the implications of nanosynaptic systems on the roadmap for nanoelectronics as set forth by ITRS (http://www.itrs.net/). The result of the analysis will be published later this year.