Startup Luminous Computing Inc., which is developing a new kind of supercomputer for running artificial intelligence models, today said that it has closed a $105 million funding round.

The Series A round included the participation of more than a half-dozen investors. The participants included Gigafund, Bill Gates, 8090 Partners, Neo, Third Kind Venture Capital, Alumni Ventures Group, Strawberry Creek Ventures, Horsley Bridge, Modern Venture Partners and others.

Mountain View, California-based Luminous is developing a supercomputer for running AI models. The supercomputer is based on silicon photonics, a technology that holds the potential to significantly speed up certain computational tasks.

A traditional processor keeps the data that it processes in the form of electricity. The processor manipulates this electricity, for example by moving it between transistors, to carry out calculations. Silicon photonics technology takes a different approach: The technology encodes data not in the form of electricity, but rather as light.

In certain circumstances, light can travel significantly faster than electricity. The vision behind silicon photonics is to leverage the fact that light travels faster in order to perform calculations more quickly than is possible with a traditional processor in which data is encoded as electricity. That’s the vision Luminous is working to implement. 

Luminous has shared few technical details about its technology thus far. However, the startup did divulge its development goals on occasion of today’s funding announcement. 

Luminous told VentureBeat that it’s aiming to build a silicon photonics chip with 3,000 times the performance of a third-generation Tensor Processing Unit, or TPU, circuit board. TPUs are specialized chips created by Google LLC for running AI models. The search giant offers the chips through its public cloud and also uses them to run internal applications.

Luminous intends to use its silicon photonics technology to build a supercomputer specifically optimized for AI workloads. The plan, according to the startup, is to harness the performance of silicon photonics to speed up AI training.

The more complex a neural network, the more hardware is required to train it. That creates challenges because the complexity of AI models is growing rapidly: OpenAI estimates that the amount of compute capacity required for the largest AI training runs doubled every three and a half months between 2012 to 2018. Luminous says its technology will make it easier to provide the growing amount of compute capacity necessary for AI development.

“Most people who build hardware assume that in order to improve performance, you have to trade off against programmability and cost-efficiency, or just go to a higher-density silicon node,” said Luminous co-founder and Chief Executive Officer Marcus Gomez. “By introducing silicon photonics technology at the heart of computer architecture, we’re not only able to drastically improve performance and scalability, but we’re also able to make it much easier to build huge AI models.” 

According to VentureBeat, Luminous has already produced multiple working prototypes of its silicon photonics chip. The startup aims to start shipping development kits to early customers within two years. According to Luminous, its initial target market consists of hyperscale data center operators such as the major cloud providers.

The startup will use its latest $105 million funding round to double its engineering team. Luminous will hire photonics designers and other experts to help advance its chip technology, as well as develop complementary software that can make it easier for future customers to use the technology. Luminous also plans to invest in scaling up chip production. 

Photo: Unsplash