Intel Unveils AI Chip that Learns To Smell, Debuts Neuromorphic System
- By John K. Waters
Researchers from Intel and Cornell University have trained a neuromorphic computer chip to recognize the scent of 10 hazardous chemicals. The chip uses an algorithm that mimics the capabilities of the "olfactory bulb" at the center of the mammalian sense of smell.
Neuromorphic chips mimic the biological structures of our nervous system. Intel's Loihi (pronounced low-ee-hee) neuromorphic chip uses an asynchronous spiking neural network (SNN) to implement adaptive, self-modifying, event-driven, fine-grained, parallel computations used for learning and inference with high efficiency. Loihi enables users to process information up to 1,000 times faster and 10,000 times more efficiently than CPUs for specialized applications like sparse coding, graph search and constraint-satisfaction problems, the company says.
The two organizations announced this capability in a paper, "Rapid online learning and robust recall in a neuromorphic olfactory circuit," published in the journal Nature, authored by Nabil Imam from Intel's Neuromorphic Computing Laboratory and Thomas A. Cleland from the Computational Physiology Laboratory in the Department of Psychology at Cornell.
"We are developing neural algorithms on Loihi that mimic what happens in your brain when you smell something," Imam said in a statement. The Loihi chips "demonstrates Loihi's potential to provide important sensing capabilities that could benefit various industries."
The developers of the Loihi processors, which were announced in 2017, took inspiration from the human brain, Intel says. "Like the brain, Loihi can process certain demanding workloads up to 1,000 times faster and 10,000 times more efficiently than conventional processors," the company said in a statement. When it was first launched, the chip provided the equivalent of 130,000 neurons and 130 million synapses. Implemented as a manycore mesh today, it now provides the equivalent of a million neurons. Each core of the chip contains a "learning engine" designed to support many types of AI models, including supervised, unsupervised and reinforcement learning.
Intel is putting the Loihi chips to work in a new experimental research system for neuromorphic computing, announced this week. Called Pohoiki (poh-ho-ee-kee) Springs, it's a 5U rack-mounted system roughly the size of five standard servers. Made up of 24 Nahuku boards with 32 chips each, the system integrates a total of 768 Loihi chips, which Intel says provides the computational capacity of 100 million neurons. The company is making the cloud-based system available to members of the Intel Neuromorphic Research Community (INRC), the members of which will be able to access and build applications on Pohoiki Springs via the cloud using Intel's Nx SDK and community-contributed software components.
Intel sees Pohoiki Springs as "the next step in scaling this architecture to assess its potential to solve, not just artificial intelligence (AI) problems, but a wide range of computationally difficult problems." Intel researchers believe the extreme parallelism and asynchronous signaling of neuromorphic systems may provide significant performance gains at dramatically reduced power levels compared with the most advanced conventional computers available today, the company said.
Intel's neuromorphic systems are still in the research phase, the company says, and not intended to replace conventional computing systems. But they provide tools for researchers to develop and "characterize" new neuro-inspired algorithms for real-time processing, problem solving, adaptation and learning.
John K. Waters is the editor in chief of a number of Converge360.com sites, with a focus on high-end development, AI and future tech. He's been writing about cutting-edge technologies and culture of Silicon Valley for more than two decades, and he's written more than a dozen books. He also co-scripted the documentary film Silicon Valley: A 100 Year Renaissance, which aired on PBS. He can be reached at email@example.com.