As AI systems develop in capability, they will almost certainly demand more computing power. This will most likely be beyond the capabilities of current computing technology. To tackle this problem, MIT spinout Lightelligence is developing next-generation computing hardware.
Compared to traditional electronic designs, Lightelligence's optical chips provide superior performance, reduced power consumption, and minimal latency. In 2017, Yichen Shen, co-founder and CEO of Lightelligence, presented a new neural network model based on particular advantages of optics in his paper "Deep learning with coherent nanophotonic circuits," which offered a two-orders-of-magnitude speed boost and a three-orders-of-magnitude power efficiency gain over cutting-edge models for learning tasks. A programmable nanophotonic processor was used to illustrate it.
The primary idea behind Artificial Neural Networks, according to the study, is modeled on the computational network models present in the nervous system. In many artificial neural network technologies, such as convolutional neural networks and recurrent neural networks, the proposed design could be used to conduct matrix multiplications and nonlinear activations.
What is the competitive advantage of Lightelligence?
Rather than utilizing the traditional fabrication platform for conventional semiconductor chips, Lightelligence employs it in a novel way. Lightelligence creates light-powered computing components that could be the backbone of the AI revolution. Lightelligence's optical processors outperform traditional architectures by orders of magnitude. Electronic circuits must combine tens, if not hundreds, of logical gates to do arithmetic. The electronic chip transistors must be turned on and off for multiple clock cycles to complete this procedure. When a logical port is turned on, it generates heat and consumes power. With Lightelligence chips, it's not the case. Shen's optic computing chips use substantially less power than their electron-powered equivalents, resulting in extremely little heat generation. Furthermore, by using just propagation, their ONN may be able to provide direct training of the network on the photonic chip with greater forward propagation speed and power efficiency.
While the CEO of lightelligence does not seek to completely replace the electronic computing sector. Rather, he wants to speed up particular linear algebra operations so that they can do fast, energy-efficient jobs like those found in neural nets.
Shen and his colleagues are not alone in this new field of optical computing. Another MIT spinoff, Lightmatter, just received an additional $80 million in Series B fundraising. Lightmatter's technology is based on a unique silicon photonics technique that uses constant light within a chip to execute calculations quickly and efficiently while using very little power. One of the contributing researchers of "Deep learning with coherent nanophotonic circuits" is Lightmatter's CEO Nick Harris. Shen and his colleagues, on the other hand, enjoy several advantages over their competitors. Lightelligence was founded in 2017 by Shen, Soljajic, and two other MIT alumni. Dr. Huaiyu Meng, Lightelligence's vice president of photonics, has a Ph.D. in electrical engineering. Spencer Powers, a business administration major, has joined the founding team. They are not only the first company to develop a complete optical hardware solution, but they also invented the technology at the institute. Shen is confident in Lightelligence's ability to innovate, regardless of competition. Lightelligence has raised approximately $40 million to far, and the team is now working on the world's largest integrated photonic system. Because data centers such as Amazon and Microsoft play a significant role in cloud-based AI computing. Centers that execute computationally expensive AI algorithms, which consume a significant amount of data center capacity. Thousands of servers that operate continually consume millions of dollars in electricity each year. Lightelligence servers, on the other side, use far less energy and are significantly less expensive. Lightelligence is a profitable startup since its AI chips not only minimize the cost but also dramatically boost processing capability.
To help their work, Newsmusk allows writers to use primary sources. White papers, government data, initial reporting, and interviews with industry experts are only a few examples. Where relevant, we also cite original research from other respected publishers.
Source- Analytics India Magazine