//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>
Luminous Computing, a California-based machine learning systems company, today announced the appointment of Michael Hochberg as President. Hochberg will lead engineering and operations at Luminous to develop what the company claims is the world’s most powerful artificial intelligence (AI) supercomputer on the market, driven by silicon photonics technology.
Hochberg brings 20 years of experience in silicon photonics to the table, having founded his first company, Simulant, at age 19, which provided the first commercial tool specifically designed for silicon photonics design, according to Luminous. . He then launched Luxtera, which claims to have been the first to manufacture silicon photonics in a CMOS foundry as well as monolithically integrate transistors with silicon photonics, and Silicon Lightwave Services. Both companies were acquired by Cisco Systems and Marlin Equity Partners respectively.
His latest venture, Elenion, commercialized 100G, 200G and 400G integrated coherent transceivers as well as 400G data center chipsets, and was acquired by Nokia in March 2020. Hochberg told EE Times that the purpose of his role at Luminous, however, is to be more of a mentor than a founder.
“A lot of people ask, why did you join Luminous? Partly it’s because it’s an adventure not starting my own startup and being more in a mentoring role. It’s a lot of fun,” Hochberg said “But a big part of why I’m here is that I think it’s the most impactful thing I can do with the skills I’ve acquired in photonics over the past 20 years. It’s really important to me that these capabilities be built in a way that is both responsible and creates asymmetric access for the United States to these technologies.
Luminous’s appointment of Hochberg as president highlights a broader vision for leveraging silicon photonics technology to deliver a purpose-built system capable of advancing AI and machine learning systems beyond their current means.
“The founders of Luminous see a huge opportunity, and that opportunity is to break down some fundamental barriers to the ability to advance the field of machine learning. Fundamentally, the world of machine learning is hardware-bound right now. The founders weren’t just looking for something to do with photonics, rather it was people who were running into these hardware limitations,” Hochberg said.
“The quality of the results you can get depends on how much computation, how much memory you can dedicate to solving these really big machine learning training problems. The field is advancing as models grow and computing hardware is getting better and better.
This is where silicon photonics becomes a key player. By placing optics at the center of computing architecture, Hochberg explains, a system can achieve computational levels beyond traditionally transmitted electrical signals and up to 1,000 times the step function in value for the end user. .
“One of the things that’s really exciting about this project is that instead of adding optics as an afterthought, where you have an electronic calculation going on and then you have optics that are used to communicate with the outside world, what we do instead is we put the optics inside the box,” Hochberg said.
“We build computer architecture around the assumption that we can build incredibly complex optical chips out of silicon and use them as the main communication path within the computer. This allows us to make architectural decisions that simply could not be made when all signals were transmitted electrically. It gives us the ability to do things that just couldn’t be done before.
It’s important to note, however, that Luminous is not an optical company; a distinction that Hochberg says is integral to distinguishing Luminous from other companies in the industry also leveraging silicon photonics.
“It is imperative to differentiate ourselves in what we do. We are a supercomputing company that does a lot of optics. We are not an optical company per se. It’s a big differentiator because it means we have a different set of customers and a different type of opportunity that we address. The machines Luminous is building are both going to have a huge revenue impact for places that use machine learning at scale today and open up these capabilities for a variety of new applications where the hardware just isn’t yet capable. to do them.
Recently released from “stealth mode,” Luminous completed a $105 million Series A round and says the funds will be used to bolster the company’s talent base, develop its custom chips and software, and prepare for a commercial scale production.