The Neuron Buddy that is 100 times faster than its biological brother

IMERI Universitas Indonesia: creating better healthcare and quality of life
August 2017
Cooperation with MIT: City Science Lab@ Taipei Tech officially launched
August 2017

The Neuron Buddy that is 100 times faster than its biological brother

Left to right: Prof Manan Suri, The Neuron Buddy and junior research scholar Thoi Singh

India – The Non-Volatile Memory Technology (NVM) Group at IIT-Delhi, led by Prof Manan Suri is in constant pursuit of cutting edge R&D in the field of neuromorphic hardware. Neuromorphic hardware corresponds to brain-inspired cognitive computing systems implemented in silicon in an energy efficient manner. In one of their recent collaborations with a Singapore-based artificial intelligence startup, Prof Suri and his team of student researchers developed what they like to call the Neuron Buddy. The Neuron Buddy is a 3D printed model – about 1000 times larger than its biological counterpart, contains more than 4000 electronic silicon neurons inside it, and works at speeds 100 times faster than the biological neuron. The entire system is powered by state-of-the-art ASIC chips, FPGAs and advanced non-volatile memory.

Shridu Verma and Narayani Bhatia, IIT-D sophomore researchers working in the group have recently demonstrated novel proof-of-concept applications of neuromorphic hardware such as multimodal biometric authentication (based on face/speech/iris recognition), and detection of abnormal cervical cancer cells from pap-smear images.

“The electronic memory as we know it, has remained a dumb ‘1’/’0’ storage for far too long, and it’s time it started taking up some share of computational tasks as our lives enter a more and more data-centric age,” says Professor Suri who is also the scientific advisor to one of the leading neuromorphic hardware companies in the world.

The group is currently developing advanced silicon architectures that exploit memory-centric computing for realising brain-inspired learning rules. Several designs which use both CMOS transistors and emerging nanoscale non-volatile memory are being explored. Going ahead, one of the key goals is to achieve low-power artificial intelligence on the edge without the need of always staying connected to large clouds.