Lightelligence claims Hummingbird is world’s first optical NoC for AI
Ahead of its first public demonstration at Hot Chips next month, Lightelligence has announced Hummingbird, believed to be the world’s first optical NoC (network on chip) processor designed for domain-specific AI workloads.
According to the company, the oNoC will revolutionise big data interconnect.
Photonic computing specialist, Lightelligence uses vertically stacked packaging for the domain-specific AI processor to integrate the oNoC and electronic ICs into a single package to serve as the communications network for data centres and other high performance applications.
Hummingbird is Lightelligence’s second photonic computing product. Its Photonic Arithmetic Computing Engine (PACE) platform integrates photonics and electronics in a small form factor leveraging custom 3D packaging and seamless co-design.
It is the first in a family of products that utilise Lightelligence’s oNOC platform, which improves computing performances by enabling innovative interconnect topologies via silicon photonics. Its waveguides propagate signals at the speed of light and utilize an all-to-all data broadcast network to each core on a 64-core domain-specific AI processor chip, for advantages in latency and power reduction over traditional digital interconnect solutions, said Lightelligence.
Unlike digital networks, Hummingbird’s oNOC technology increases density scaling by enabling interconnect topologies that would be otherwise unrealisable, said the company.
In oNOC, power and latency are virtually unaffected by distance, making the technology suitable for developing robust topologies that do not rely on nearest neighbour communication. oNOC topologies like Hummingbird’s enable higher utilisation of compute power even in a single electronic IC configuration due to more efficient communication. With oNOC, mapping workloads to hardware becomes easier and provides greater freedom to select the right topology for the computing task.
In Hummingbird, Lightelligence implemented a low-latency optical all-to-all broadcast network spanning 64 cores. With 64 transmitters and 512 receivers, Hummingbird provides a framework to implement a variety of dense optical network topologies.
The electronic and photonic ICs are co-packaged and integrated into a PCIe form factor ready for installation in industry-standard servers. Lightelligence also offers a software development kit (SDK), machine learning and AI workloads can be optimised to take full advantage of the oNOC. oNOC and Hummingbird IP can also be customised for other workloads and applications.
Future generations of Hummingbird will employ reticle-stitching to support chiplet architectures to enable better scalability, improve energy efficiency and further reduce bottlenecks.
Lightelligence is actively signing development partners to sample Hummingbird-based PCIe add-in cards along with Lightelligence’s SDK in Q3 2023.
The first public demonstration of Hummingbird will be at Hot Chips August 27-29 at Stanford University, California, USA. Lightelligence will also exhibit before Hot Chips at the Flash Memory Summit at the Santa Clara Convention Center in Santa Clara, California (8 to 10 August 2023)
Confirming his belief in the value of photonics computing, Lightelligence’s CEO, said: “Photonics is the solution to the critical compute scaling problem, which has become pressing as the traditional solutions struggle to keep up with the exponential growth of compute power demand spurred by breakthroughs in the AI industry,” he said. He added that Hummingbird can address the scaling problem by incorporating photonic technologies into next-generation products.
“Lightelligence is breaking the memory wall with its proprietary photonics technology that could revolutionise the semiconductor industry,” confirmed Dylan Patel, chief analyst at SemiAnalysis.
Lightelligence operates in the photonic technology space to offer exponential improvements in computing power and dramatically reduce energy consumption. It is the only company that has publicly demonstrated integrated silicon photonic computing systems working at speed.
