Intel Smells Neuromorphic Opportunity

A photo shows Intel’s latest neuromorphic system, Pohoiki Springs, and one of the rows within it. The system unveiled in March 2020 integrates 768 Loihi neuromorphic research chips inside a chassis the size of five standard servers. (Credit: Intel Corporation)

Neuromorphic computing has a rather long way to go before it becomes an accepted part of systems. Like its quantum brethren, mapping problems to the architecture is still a heady challenge even though a few use cases show remarkable promise. Further, just as with quantum computing, most major chip and systems players have some interest in exploring the technical possibilities and Intel is no exception.

Since the scope of problems that can be tackled with neuromorphic chips is still limited, Intel focused on one very specific use case to highlight the advance from its last 64-chip system based on its “Loihi” architecture. Using a system called “Pohoiki Springs” that has scaled to 768 Loihi chips (100 million spiking neurons) sitting in a 5U rack-mount chassis Intel showed how a neuromorphic system can pick out smells with a small training sample, high accuracy, and in a system that consumes only 300 Watts.

When we first described “Pohoiki Beach” which is based on Intel’s “Nahuku” boards, each of which contained eight to 32 Loihi processors, it was a 64-processor system built from between two and eight boards (Intel did not offer details of the exact configuration, including how the chips and boards are networked together (and still has not). That same system has now been scaled to the aforementioned 768 chips.

Loihi, which Intel unveiled in 2017, initially provided the equivalent of 130,000 neurons and 130 million synapses, implemented as a manycore mesh, a dramatic increase now in 2020 with over one million neurons.  Each core contains a “learning engine” that can support different many types of AI models, including supervised, unsupervised, and reinforcement learning, among others. According to Intel, Loihi is about 1,000 times faster and 10,000 times more efficient than CPUs for applications like sparse coding, graph search and constraint-satisfaction problems. The chip has been available to researchers through the Intel Neuromorphic Research Community (INRC) via a cloud service and as the Kapoho Bay platform, a Loihi-based USB form factor device. (And yes, if Intel wanted to work toward commercializing an esoteric technology, it could have done the world a favor and not done it all under names that aren’t sticky either).

Like the brain, Loihi can process certain demanding workloads up to 1,000 times faster and 10,000 times more efficiently than conventional processors. Pohoiki Springs is the next step in scaling this architecture to assess its potential to solve not just artificial intelligence 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.

According to Mike Davies, who heads Intel’s neuromorphic computing program, these specialty systems can be used by doctors to sniff out diseases, in airports to detect weapons, drugs, or bombs, or dangerous chemicals at manufacturing sites, for instance.

Although it might sound like a stretch to use a specialized neuromorphic architecture with a fussy programming suite to do all of this when a neural network could also pick up similar patterns (as Google and others have shown) there are some features of a neuromorphic system that traditional deep learning models and machines can’t touch. The energy efficiency and time to result are the two most prominent.

The efficiency gains come from fully integrating compute and memory on a neuromorphic system. There is no separate memory that streaming instructions and data need to swing through. Everything is integrated into one distributed fabric of compute and memory. As Davies explains, it all boils down to asymmetry. “Wanting the system to communicate or not is a power expenditure question. If you don’t send anything that’s a 0 binary value; not sending means not using energy. Coding information in this temporal way, sending at a point in time can encode is a way to send info and you can compute with those codes in a way that allows you to prefer a “0” state.” The problem is that getting to that state requires a rethink of algorithms. This is what the spikes are all about in a “spiking” neuromorphic system, he adds.

“Pohoiki Springs scales up our Loihi neuromorphic research chip by more than 750 times, while operating at a power level of under 500 watts. The system enables our research partners to explore ways to accelerate workloads that run slowly today on conventional architectures, including high-performance computing (HPC) systems,” says Davies.

Sign up to our Newsletter

Featuring highlights, analysis, and stories from the week directly from us to your inbox with nothing in between.
Subscribe now

Be the first to comment

Leave a Reply

Your email address will not be published.


This site uses Akismet to reduce spam. Learn how your comment data is processed.