NEWS  – A breakthrough in neuromorphic computing was reported by scientists from IISc Bengaluru

HIGHLIGHTS

• The research could propel India into the global AI race, with potential to revolutionize AI computing.
• This development moves away from conventional cloud computing to a more energy-efficient edge computing

Key Discovery: Memristor Technology

• The IISc team, led by Prof Sreetosh Goswami, developed a Memristor using a metal-organic film, moving away from conventional silicon-based technology.
• The Memristor mimics the brain’s neural networks, processing information like neurons and synapses rather than in binary (0s and 1s).

Importance of Memristor in AI

• When integrated with conventional computers, the Memristor boosts performance by hundreds of times, making it a highly energy-efficient AI accelerator.
• The team demonstrated energy efficiency of 4.1 tera-operations per second per Watt (TOPS/W), 220 times more efficient than the NVIDIA K80 GPU.

Neuromorphic Computing vs. Traditional Computing

• Traditional computing using Von Neumann architecture is running into speed and energy efficiency limits, especially with AI’s growing demands.
• Neuromorphic computing, inspired by the analog brain, processes large chunks of data in fewer steps, making it faster and more energy-efficient.

Applications and Demonstrations

• The Memristor, integrated with a desktop computer, successfully processed complex space images from NASA’s “Pillars of Creation” image.
• The device can process 16,520 states simultaneously, reducing operations for AI algorithms like matrix multiplication

Path Forward

• The team demonstrated a 64X64 Memristor array and plans to scale up to 256X256 arrays.
• With funding from MEITY, the team aims to develop a System-on-Chip solution and incubate a startup for commercial application within the next three years.