#307 Steven Brightfield: How Neuromorphic Computing Cuts Inference Power by 10x

What It Covers

BrainChip's Steven Brightfield explains how neuromorphic computing chips reduce AI inference power consumption by 10x through event-based processing inspired by biological neurons, enabling always-on edge AI in wearables, autonomous systems, and consumer devices.

Key Questions Answered

• •Power efficiency breakthrough: Neuromorphic chips process only data changes rather than continuous streams, reducing power by 10x compared to traditional AI processors. This extends battery life from one month to one year in doorbell cameras and enables week-long smartwatch operation.
• •Event-based processing advantage: Unlike GPUs that compute every pixel every thirtieth of a second regardless of activity, neuromorphic chips generate spikes only when changes occur in sensor data. A blank camera screen consumes almost no power until movement triggers computation, eliminating wasted calculations on zeros.
• •Commercial production milestone: BrainChip enters volume production with GlobalFoundries manufacturing 22-nanometer neuromorphic chips in Upstate New York. The company offers both chips and IP licensing, allowing customers to validate performance before committing tens of millions to custom chip development.
• •Edge AI applications expanding: Neuromorphic chips enable sub-millisecond object detection for autonomous vehicles, epileptic seizure prediction in smart glasses, and radar-based object classification for indoor drone navigation. Market research projects 35% of embedded devices will run AI within five years, with neuromorphic in half.