Alladi Ramakrishnan Hall

Molecular hardware for Artificial Intelligence

Sreetosh Goswami

Centre for Nano Science and Engineering, Indian Institute of Science, Bengaluru

Artificial Intelligence (AI) is a concept that has seen many hype cycles arrive and fade as we have become overly impressed by our computing machines and then dismayed by our realization of the complexity of what brains actually do. In recent years, we might believe that the age of AI has finally arrived as algorithms beat human champions at complex games and the champions retire in frustration. However, a look into the backroom reveals that the machines that run these algorithms require kilowatts of power to operate, and the amount of preparation and training of the algorithms is long and extremely costly. When it comes to functions like intelligence, cognition, and decision making, any of today's best computing systems is outperformed by the brain by orders of magnitude both in terms of energy and space efficiency. The primary reason for our shortcoming is that we are using conventional passive circuit elements to emulate a highly non-linear dynamic operating system of a brain that operates on the verge of chaos. In this talk, I shall introduce a new generation of molecular circuit elements that can capture complex, reconfigurable, dynamic logic in nano-scale material properties and can also be poised on the verge of instability. These devices could be optimal for emulating a brain's functions.