Editor's note:“Some people are skeptical about the investment of chip start-ups because it is a capital-intensive industry.”
Boston's Lightmatter is developing a chip with optical components that can avoid the limitations of the current generation of chips. In response, GV, a venture capital firm owned by Alphabet, the parent company of Google, has invested. At present, the specific time for the new chips to go on the market is unknown.
A small start-up in Boston had a strange idea of developing a fast and efficient processor for AI that would work as long as the light was shining.
More specifically, Lightmatter's chip includes an optical element called the Mach-Zehnder interferometer, rather than a more common accumulator or MAC unit. This swap is designed to circumvent the limitations of chips currently on the market.
Lightmatter, the venture capital arm of Alphabet, the parent company of Google, has just received its first support from GV. This indicates that there is a new type of hardware that can accelerate the expansion of the development space of AI. At present, AI is attracting more and more attention in many fields besides retail and health care technologies.
When investors first heard the idea, some people were skeptical. Thomas Graham, co-founder and chief operating officer of Lightmatter, recalled the mood at the time.“Their suspicions have taught me that it's actually more than just a science project.”But now the team has been recognized by one of the world's top technology companies.
The Way to End Moore's Law
In 2014, Nick Harris and Darius Bunandar tried to combine optical technology with quantum computing at MIT, when they were working on a PhD in the same research group.
Quantum computing is different from traditional computing, which can be reduced to simple 1 and 0. Quantum computing involves the creation of quantum bits, or quantum bits, which can be expressed as 1 and 0 at the same time. Quantum computers can perform certain types of computing faster than traditional computers, and Google, IBM, Intel and Microsoft are exploring this area.
But in 2015, Harris and Bunandar began to focus on areas other than quantum computing, including artificial intelligence. Harris said,“We believe that there are still many difficult challenges for ordinary quantum computing methods.”。
Both are engineers, so they know little about business. So they decided to study entrepreneurship at MIT Sloan School of Management. There, they met Graham, a former Morgan Stanley analyst. The course requires students to come up with business ideas and make videos about them. When Graham saw Harris's video of computing with light, he appreciated it.
So the three decided to compete in the MIT $100,000 entrepreneurship contest. Harris representation“This is a good thing for our class.”They took the game seriously and worked out a business plan. In the meantime, they became friends.
They eventually defeated dozens of other teams, won the 2017 competition and won a prize of $100,000. However, one of the judges in the competition happened to be Erik Nordlander, GV's general partner, who shook hands and talked to them after the trio's victory. Later, the three went to Silicon Valley with their bonuses to meet with more investors. Meanwhile, a paper on their unconventional methods was published in Nature Photonics.
In Silicon Valley, these three people become more intimate and live and eat together. They also brought chips developed at MIT to meetings with venture capitalists. Not long ago, however, these investors were reluctant to inject capital into semiconductor companies.
Harris said:“Some people are skeptical about the investment of chip start-ups because it is a capital-intensive industry.”
Even so, Graham said, investors they met were aware of the challenge of Moore's law, that engineers could double the number of transistors on a single chip every two years.
The speech also pointed out that companies using optical technology to achieve rapid networking in data centers are rising. In earlier years, investors injected large amounts of money into these companies, such as Applied Optoelectronics.
As investors poured in, the Lightmatter team began to work. In a year, it produced more than one, but two early chips. The latest one contains more than one billion transistors.
Now, GV has signed a partnership agreement with Spark Capital and Matrix Partners, providing the startup with $22 million in new capital. The company's founders are excited about working with GV to recruit people to develop software for chips.
Harris said:“We've spent a lot of energy building this plug and play device to make it look more like Nvidia GPU.”The team wants to ensure that chips can be used with popular AI software, such as TensorFlow, an open source project supported by Google.
The initial focus of the project was to sell chips to organizations running large cloud computing data centers and high performance computing clusters. At the same time, other start-ups, such as Cerebras and Graphcore, are developing chips that can be used to model AI in these places, and large companies, including Intel, want a share.
But Harris and his team believe they will have an advantage. In data centers, he says, two things are critical: the number of operations that can be performed per second or production, and the number of operations that can be performed at efficiency or per unit power.
He said:“Our system will be able to provide 10 times better solutions than existing solutions.”
‘Point the day and await for it’
The biggest question now is how long it will take Lightmatter, with 23 employees, to release its first batch of chips.
Harris representation“This is definitely not a long-term, five to ten-year event. This will happen soon.”But he did not provide a more specific timetable.
GV has only one limited partner to fund it.——Alphabet, Google's parent company. As a result, the company can make longer-term bets without worrying about returning funds to its limited partners.
Tyson Clark, GV's general partner, said:“We do have more patient capital than other ordinary venture capital.”
Clark says there are risks in silicon photonics because it is not as stable as traditional semiconductor technology. But he said that as the world began to be constrained by standard computing systems, the demand for chip computing power would only increase in the future.
He said he would be happy to introduce Lightmatter to help staff on Google's public cloud. Moreover, he did not have any obstacles when talking about the start-up company with employees of large cloud services companies such as Amazon Web Services and Microsoft Azure.
Clark said:“I hope this company will succeed and become the next Intel or Nvidia, and I will use all my relationships anywhere to help them achieve this goal.”