Is neuromorphic the best architecture for future SoCs?
Approaching power/performance tradeoffs from an architectural perspective is essential given the complexities of today’s SoCs. And beyond some traditional techniques that I discussed in a recent article, Bernard Murphy, CTO at Atrenta mentioned that there is currently a lot of buzz about using non-Von Neumann architectures — especially for recognition functions (voice, image and text).
He said a leading architecture option is neuromorphic computing, which can perform as well as or better than traditional architecture at three orders of magnitude less power. Murphy feels that given increasing focus on biometric IDs and voice control, this is likely to become a big contributor to freeing up performance options for other functions in the design.
IBM is working on this, along with Rensselaer Polytechnic Institute, Stanford, and Qualcomm, along others.
There is even a neuromorphic computing platform being developed, and last month, Nanyang Technical University in Singapore revealed details of a new technology they think is already capable of ‘brain-like’ computing.
Where these interesting developments will lead to is still to be seen, but there is clearly much to be gleaned from the technology and how to apply it to SoC design.
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Leave a Reply