British Photonic Computing Startup Raises Seed Round

1652345917 Salience chip close up

//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>

British photonics computing startup Salience Labs has raised an $11.5 million funding round to develop its hybrid photonics-electronics chip, which will ultimately target accelerating AI inference in applications requiring low latency. This includes robotics, vision systems, healthcare, and many other applications.

Salience, a spin-out from the universities of Oxford and Münster that was created in 2021, is aiming for an order of magnitude increase in performance over current microchips with its first product.

The concept of photonic computing isn’t new, but the supporting technologies required are just coming together, said Salience Labs co-founder and CEO Vaysh Kewada. EE time.

Vaysh Kewada (Source: Salience Labs)
Vaysh Kewada (Source: Salience Labs)

“The reason we think it’s possible today is because the manufacturing process has come a long way in the last five to seven years,” she said. “It is now possible to manufacture a photonic chip using CMOS processes in a production foundry; you can go tomorrow and tape a photonic chip. The development of this manufacturing process was essentially based on the development of optical transceivers, and what we do at Salience Labs is to co-opt these components that already exist to build a photonic processor.

Salience Labs will innovate in both photonics and electronics. On the photonics side, the company will build on the work done by Salience co-founder Johannes Feldmann during his PhD. He built a prototype photonic computer chip that encodes data using the amplitude of light, then modulates that light to perform matrix multiplication at extremely high speeds. The computing element uses an electro-optic modulator, applying a voltage to the waveguide to modulate the light.

READ MORE:  Explore liquid cooling for data centers

Compared to other optical phase-based approaches, coherent light is not required for the Salience amplitude-based approach, and there are no interferometers, which simplifies the architecture , said Feldmann. EE time. Until recently, this amplitude-based approach relied on components that couldn’t be made in a foundry, but that’s changed, Kewada said.

“At the time when [other photonic computing companies] were established, you would have looked at the amplitude-based approach and thought, that’s great, but you can’t make that in a foundry,” she said. “One of the things we’ve done at Salience is move towards fully foundry-based components, and that step happened after everyone else was established.”

The advantages of Salience’s amplitude-based approach include the ability to clock the chip to tens of gigahertz and the ability to easily perform multiple calculations simultaneously using different wavelengths of light (up to 64 vectors can be stacked using this technique). The result, Kewada said, is super-high-bandwidth computing suitable for accelerating low-latency AI.

READ MORE:  TACC's Stampede2 Supercomputer Simulates Water Supply in Inter-utility Agreement Study
Johannes Feldmann (Source: Salience Labs)
Johannes Feldmann (Source: Salience Labs)

A paper on Feldmann’s prototype chip, published in Nature, describes how the prototype chip was used to run convolutional neural networks for image recognition, demonstrating matrix multiplication up to 13 GHz and multiplexing with four d-lengths. simultaneous waves of light.

Salience is also working on a partner electronic ASIC for data orchestration. The ASIC chiplet has a highly distributed memory architecture that allows data to be pumped into the photonic chiplet, keeping it heavily utilized.

“How to power a chip that syncs much faster than electronics is at the heart of our innovation,” Kewada said. “It’s about how we stack our photonics chip on top of our memory in a sort of in-memory computing architecture.”

Salience’s photonic chip will be stacked on top of its electronic chip, keeping the chips as close together as possible using existing packaging technologies, most likely an interposer-based solution, Kewada said.

Salience Labs is up against more established startups including Lightmatter and Lightelligence in this space. Kewada said there is more than enough room for multiple photonic computing architectures to coexist.

Photonic Computing Electro-Optical Assembly from Salience Labs
A rendering of the Salience electro-optical assembly (Source: Salience Labs)

“Going forward, there will be a fairly large amount of compute workloads in a fairly large amount of segmentation, and huge amounts of different hardware that will target these different use cases,” she said. declared. “At Salience, we focus on the areas where our technology really shines: the use cases where we know we can add value.”

READ MORE:  CLOU Stock: Growth Is Fully Reflected In Valuations (NASDAQ:CLOU)

The hybrid Salience chip will initially target inference workloads (although training workloads are possible, Kewada said), especially those that can take advantage of the design’s low latency. This includes AI-based communication signal processing in 5G base stations, robotics, vision systems and more.

Salience photonic computing chip (Source: Salience Labs)
Salience photonic computing chip (Source: Salience Labs)

“We have great confidence in our photonics approach, which gives us significant computational throughput and allows us to continue to scale,” she said. “The beauty of a hybrid photonics-electronics approach is that you can scale it up and go far beyond [our current prototype] in a way that electronic architectures cannot.

Salience’s seed funding round was led by Cambridge Innovation Capital and Oxford Sciences Enterprises, with participation from Oxford Investment Consultants, former Dialog Semiconductor CEO Jalal Bagherli, former board member from Temasek Yew Lin Goh and Arm-backed Deeptech Labs.

Salience Labs now has around 10 employees, most of whom are in the UK, and the company aims to grow its workforce to around 15 by the end of the year.

Source link

Leave a Comment