Optical computing startup Lightelligence has just showcased the computing power of its new chip device. It can solve the Ising problem (a mathematical model of ferromagnetism in probabilistic mechanics, a first-order fit model in understanding conversions between forms) 100 times faster than a single system. system with GPU-based processing power.
According to the developer team, the chip “Hundreds of times faster than other common computing devices, such as NVIDIA RTX 3080“.
Called PACE (short for Photonic Arithmetic Computing Engine), the optical computing device contains 12,000 photonic devices running at 1 Ghz. Compared to the experimental machine COMET, which owns 100 photonic devices, launched by Lightelligence in 2019, the new machine is about 1 million times faster.
Speaking to EETimes, Yichen Shen, CEO of Lightelligence emphasized: “The heart of our optical computing engine is that it completes matrix multiplication in much less time. [một GPU]“. According to the company’s claim, a GPU can use up to hundreds of operations to complete a 64×64 matrix multiplication, while PACE can do it in as little as 10 or 5 nanoseconds.
Thanks to signals traveling at speeds close to light, PACE’s latency – or the time it takes for a machine to successfully execute an instruction – is very low. To be successful, Lightelligence had to focus on both the capabilities of optical technology and the power of traditional transistors, in addition to software capable of linking the two factors mentioned above.
For that reason, Lightelligence claims to be a company that develops both hardware and software. They design algorithms to solve the most basic computing problems, but in photovoltaic environments.
“One of Lightelligence’s most unique advantages over other optical computing developers is its ability to design for multiple perspectives at once.“, said engineer Erwan Di Vita working at Lightelligence. “Our photovoltaic engineers collaborate with [chuyên gia từ nhiều mảng khác] to design the chips, and then our post-silicon team will manufacture them and build them into 3D systems“.
As commented by Yichen Shen, professor and founder and CEO of Lightelligence, commented: “These problems fall under the category of difficult math problems, known as NP-complete, that have baffled mathematicians for the past 50 years.“.
CEO Shen continued: “Algorithms for NP-complete problems are important because they […] There are hundreds of practical applications in areas including cryptography, grid optimization or advanced image analysis. The advances we have made with the NP-problem combinatorial optimization problem show the technology’s potential to transform the computing industry.“.
Optical computing based on silicon photovoltaics can increase computer processing speed and performance many times over. However, we need to clearly distinguish the capabilities of the two devices running on or different mechanisms. Photovoltaic devices are ideal for applications in linear computing operations such as matrix multiplication, while digital electronics are commonly found in non-linear operations, besides control. and save data.
Difficulties still piled up before Lightelligence. According to Maurice Steinman, the company’s vice president of engineering, good results will be needed “a great deal of effort is poured into circuit design, simulation, and repeated testing of the chips“. In addition, there are difficulties in combining two types of technology into a single device.
“Something that works with light, so we need to put a light source in it. The other device needs current to operate … again to consider heat dissipation. There are a lot of obstacles that need to be addressed in an orderly fashion so that they all come together” concluded Mr. Steinman.
Lightelligence has launched its first commercial product, a system-accelerated AI system that runs on the technology that makes up the PACE appliance. It is expected that the startup will ship the first finished products in 2022. Up to now, Lightelligence has raised $ 100 million in capital, owns 150 employees from many parts of the world.