Sounds a little flashy but the company's team and technology are real: Nick Harris, Lightmatter's chief executive, wrote about photonic chips in a MIT paper and posted a number of articles on photonic computing architecture feasibility.
The company is located in:
Like most computational operations, AI and machine learning are fundamentally computationally based on a large number of calculations. Ordinary computers can do simple calculations, but for complex problems, they need to be disassembled and broken down, each broken. The problem, which is common in AI applications, is that the more crucial question is how to get AI to handle large data sets quickly.
As the demand for artificial intelligence products continues to increase, people are demanding more and more computing, but it is clear that the current computer chip has reached its limit. "One of the defining demise of Moore's Law is that companies like Intel are Invest heavily in things like quantum computing, invest in development as long as it is not traditional computing, "Harris said in an interview." So now is a great time to look for other computing architectures. "
The company's photonic chip solves all of a sudden problem by letting light pass through a tiny configurable lens and sensor, and by creating and tracking small changes in the phase or path of light, only when light goes from one end of the chip to the other The solution can be generated in extreme times, which means not only quick feedback of the calculation results, but also the energy required to consume a very small part of the traditional chip.
'A lot of deep learning relies on this particular run, and our chip can greatly increase that speed,' explains Harris. 'This is a special-purpose optical computer that can shine on special issues. The first photonic chip can accomplish this task in a precise and scalable way. '
'We are very close to the theoretical possibilities in terms of speed, power consumption and latency,' says Harris, meaning you can not get the light to run faster, but you can put the chips in a much more dense place, like a traditional computer, Let them work in parallel, or, improve sensor performance and more.
The company recently announced a $ 11 million Series A round of funding led by Matrix and Spark, a new fund that will help teams apply technology from the prototype level to real products.
'This is not a science project,' said Stan Reiss of Matrix, 'This is the first application of optical computing in a controlled way.'
He also pointed out that competitors are too rigidly focused on improving the performance of semi-specialized hardware, such as the development of special AI motherboards, but eventually they can not jump beyond the limitations of traditional computers.
'Anyone can develop this chip, but the problem is that such chips are also facing a great competitive pressure, while Lightmatter this company is completely different.' And, photon chips have only recently become possible. The investment in basic research and infrastructure to build photonic chips has paid off in the past ten years, and now the technology is mature enough to move from the lab to the market. "Harris said: 'AI is still in its infancy, Moving forward, new technologies are essential. At Lightmatter, we are using photonics to enhance the performance of electronic computers to radically evolve powerful enough new computers to advance the next generation of artificial intelligence. "