Ayar Labs to Demo Photonics Chiplet in FPGA Package at Hot Chips

By Tiffany Trader

August 19, 2019

Silicon startup Ayar Labs continues to gain momentum with its DARPA-backed optical chiplet technology that puts advanced electronics and optics on the same chip using standard CMOS fabrication. At Hot Chips 31 in Stanford, Calif., this week, Ayar Labs will show results of the work it’s been doing with DARPA and Intel on an FPGA chiplet integration platform targeting radar applications.

Ayar Labs, which has offices in Emeryville and Santa Clara, Calif., has just taped out its 11th generation “AL11” electrical-to-optical I/O chiplet, the basis for its commercial product, called TeraPHY, which the company says will enable optical I/O bandwidth capability in excess of a Terabit/sec at 10X lower power than today’s Gigabit/sec I/O solutions.

TeraPHY will be integrated with other silicon technology partners’ designs into multi-chip module (MCM) computing products. In a stroke of good timing for Ayar Labs–which expects early system deployments and prototypes in “reasonable volume” in 2020–there’s something of an MCM/chiplet revolution afoot. Chipmakers, such as Intel and AMD, are turning to the modular packaging approach to sidestep Moore’s law (or maybe Gordon Moore actually foresaw this).

Ayar Labs has been working closely with Intel’s Programmable Solutions Group as part of the DARPA CHIPS (Common Heterogeneous Integration and IP Reuse Strategies) project. At Hot Chips this week, Ayar Labs will be showing its 10th gen chiplet installed in an Intel FPGA package with an Intel Stratix 10 die; and it will also announce its support for the Advanced Interface Bus (AIB) protocol.

ERI Summit (July 16, 2019)

Intel’s Senior Principal Engineer Sergey Shumarayev previewed the new technology at DARPA’s ERI Summit last month. The Ayar Labs chiplet connects to the Intel Stratix 10 FPGA die via the AIB interface using Intel’s EMIB packaging. The Jariet Technologies’ millimeter wave chiplet gets information in through RF, passes that information to the FPGA for pre-processing and that gets shipped out optically via the Ayar Labs chiplet.

The University of Michigan chiplet functions as an AI accelerator. Ayar Labs is not directly involved in the University of Michigan research project, but Ayar Labs Chief Scientist and Co-Founder Mark Wade noted the AI chiplet illustrates the versatility of this type of integration.

“This chiplet package ecosystem allows someone to put in an AI accelerator chip and interface to the FPGA, and then also on the other side interface to our high performance optical IO chip; there’s a kind of transformative joint solution coming together that can be enabled by this emerging ecosystem,” said Wade, who will be presenting at Hot Chips on Tuesday.

Ayar Labs has been working with Intel to evolve AIB — the wide, parallel PHY-level interface that Intel open sourced to help spur chiplet adoption. Freely available on GitHub, AIB is being driven as an open industry standard for enabling low-latency, power-efficient direct chiplet-to-chiplet communication.

Ayar Labs also continues to work on support for serial interfaces, like SerDes — an approach of course that has a lot of momentum behind it. Wade said Ayar’s roadmap now includes a provision for both wide parallel implementations, such as AIB, as well as serial implementations, but he says they are getting the most traction with the wide parallel approach.

Ayar Labs CEO Charlie Wuischpard, who joined the company in November 2018, after heading HPC at Intel for a number of years, told HPCwire that while the FPGA solution being demoed this week is emerging as an Intel technology, the opportunity extends beyond working with a single vendor.

“The fact that the government is standardizing on an electrical interface makes this really suitable for anybody else to pick up and run with,” he said. “We’ve been getting a ton of interest from across the industry as there’s this greater recognition that optical I/O creates whole new sets of opportunities.

Example of a possible heterogeneous system in package (SiP) that combines sensors, proprietary ASIC, FPGA, CPU, Memory, and I/O using AIB as the chiplet interface. (Source: Intel Corp.)

“One of the really interesting things about our technology is not just the innovative optical chiplet that goes in the package, but once you’re relieved of the constraints that are currently in place in system designs around the power and packaging and cooling to drive the massive signals, and that’s the bandwidth over copper, then you can create brand new system architectures,” said Wuischpard.

The FPGA demonstration targets phased array radar applications, which is an important market target for Ayar Labs along with 5G, but Hugo Saleh, vice president of marketing and business development, who also made the jump from Intel, told HPCwire they anticipate another killer app: high-end HPC and AI. Picture, for example, an Nvidia DGX-2 type-box, but instead of the NVLinks, there would be 16 GPUs or AI chips connected by optical interfaces, and then multiple boxes could be connected using optical instead of switched fabrics.

Potential use cases extend to any server with a PCIe card in it. “We’ve got a prototype design where essentially you could take a GPU or an FPGA or an ASIC and build a system the size of a whole rack or maybe multiple racks all optically connected that far expands beyond the 16 GPUs currently supported [in the DGX2],” said Saleh.

The glueless, point-to-point interconnect approach applies for standard two-socket servers, he added. Ayar Labs foresees using optical to create supernodes where every element of the system can talk to every other element of the system, versus going up through a switch.

“We’ve reviewed this with several of the large hyperscale providers and the day we have this capability, they want to start testing it,” said Wuischpard. “We have to be able to first produce our single instance of optically connected FPGA as a first step, but we very rapidly want to move to these new system designs that use optical I/O in a new way.”

AI applications and graph analytics are clamoring for this kind of memory semantic type architecture, and government agencies are looking to leverage the technology for advanced threat intelligence, for identifying patterns in vast and disparate data sets and for tracking objects—like airplanes and drones-in real-time.

Ayar Labs has in its pipeline a graph analytics design win that is using four Ayar Labs chiplets per socket with each one of those chiplets putting out 1.3 Terabits per second of data. “That’s an aggregate of over 5 Terabits per second out of package and the only way they can do this is with optical,” Saleh stated.

In the lab: co-packaged Stratix 10 FPGA and TeraPHY chiplets with fiber connected (courtesy: Ayar Labs)

Ayar Labs works closely with GlobalFoundries, a trusted foundry for the U.S. government. The TeraPHY chips are being manufactured using GloFo’s 45 nanometer RF SOI (Radio Frequency Silicon on Insulator) process, currently at its East Fishkill, NY, facility.

Each optoelectronic chip spans an area of approximately five by eight millimeters. Wade said the sizing takes into account how the chips line up with the Intel FPGA. The Intel chip is a full reticle FPGA die, about 25 millimeters on one edge, and the entire east-west side of that edge has the AIB interface on it. That allows up to three Ayar dies to abut each of those edges, to connect up to the AIB interface and escape all that bandwidth.

“This chiplet kind of revolution that is happening right now really allows a mix and match type of configurability that was not there before,” said Wade. “So it allows you to essentially customize at the package level a chip solution that can be targeted to specific applications. In some cases, you might use an FPGA that doesn’t need every slot populated with an optical I/O chip. So maybe you have a memory chip, maybe you have an AI accelerator chip, maybe you have one of our chiplets. And the beauty is you can mix and match all these things together in a heterogeneous way that previously really was not possible. So this emergence of the chiplet ecosystem is really a big development enabling our type of solution.”

Ayar Labs was founded in 2015 by a group of researchers from MIT, UC Berkeley, and CU Boulder who were part of a 10-year research collaboration funded by DARPA. The company has gone through several funding rounds with an investor roster that includes Playground Global, Intel Capital, Global Foundries, the Founders Fund, Silicon Valley Bank and others. It has also been the recipient of several recent DARPA grants and is involved in the CHIPS and PIPES projects.

Feature image caption: a cameo shot in the lab of Ayar Labs’ TeraPHY chiplet co-packaged with a Stratix 10 FPGA (courtesy: Ayar Labs)

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion XL — were added to the benchmark suite as MLPerf continues Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing power it brings to artificial intelligence.  Nvidia's DGX Read more…

Call for Participation in Workshop on Potential NSF CISE Quantum Initiative

March 26, 2024

Editor’s Note: Next month there will be a workshop to discuss what a quantum initiative led by NSF’s Computer, Information Science and Engineering (CISE) directorate could entail. The details are posted below in a Ca Read more…

Waseda U. Researchers Reports New Quantum Algorithm for Speeding Optimization

March 25, 2024

Optimization problems cover a wide range of applications and are often cited as good candidates for quantum computing. However, the execution time for constrained combinatorial optimization applications on quantum device Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at the network layer threatens to make bigger and brawnier pro Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HBM3E memory as well as the the ability to train 1 trillion pa Read more…

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HB Read more…

Nvidia Looks to Accelerate GenAI Adoption with NIM

March 19, 2024

Today at the GPU Technology Conference, Nvidia launched a new offering aimed at helping customers quickly deploy their generative AI applications in a secure, s Read more…

The Generative AI Future Is Now, Nvidia’s Huang Says

March 19, 2024

We are in the early days of a transformative shift in how business gets done thanks to the advent of generative AI, according to Nvidia CEO and cofounder Jensen Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. Wh Read more…

Alibaba Shuts Down its Quantum Computing Effort

November 30, 2023

In case you missed it, China’s e-commerce giant Alibaba has shut down its quantum computing research effort. It’s not entirely clear what drove the change. Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Shutterstock 1285747942

AMD’s Horsepower-packed MI300X GPU Beats Nvidia’s Upcoming H200

December 7, 2023

AMD and Nvidia are locked in an AI performance battle – much like the gaming GPU performance clash the companies have waged for decades. AMD has claimed it Read more…

DoD Takes a Long View of Quantum Computing

December 19, 2023

Given the large sums tied to expensive weapon systems – think $100-million-plus per F-35 fighter – it’s easy to forget the U.S. Department of Defense is a Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Intel’s Server and PC Chip Development Will Blur After 2025

January 15, 2024

Intel's dealing with much more than chip rivals breathing down its neck; it is simultaneously integrating a bevy of new technologies such as chiplets, artificia Read more…

Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move

January 5, 2024

Reuters reported this week that Baidu, China’s giant e-commerce and services provider, is exiting the quantum computing development arena. Reuters reported � Read more…

Leading Solution Providers

Contributors

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Shutterstock 1179408610

Google Addresses the Mysteries of Its Hypercomputer 

December 28, 2023

When Google launched its Hypercomputer earlier this month (December 2023), the first reaction was, "Say what?" It turns out that the Hypercomputer is Google's t Read more…

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

Shutterstock 1606064203

Meta’s Zuckerberg Puts Its AI Future in the Hands of 600,000 GPUs

January 25, 2024

In under two minutes, Meta's CEO, Mark Zuckerberg, laid out the company's AI plans, which included a plan to build an artificial intelligence system with the eq Read more…

Google Introduces ‘Hypercomputer’ to Its AI Infrastructure

December 11, 2023

Google ran out of monikers to describe its new AI system released on December 7. Supercomputer perhaps wasn't an apt description, so it settled on Hypercomputer Read more…

China Is All In on a RISC-V Future

January 8, 2024

The state of RISC-V in China was discussed in a recent report released by the Jamestown Foundation, a Washington, D.C.-based think tank. The report, entitled "E Read more…

Intel Won’t Have a Xeon Max Chip with New Emerald Rapids CPU

December 14, 2023

As expected, Intel officially announced its 5th generation Xeon server chips codenamed Emerald Rapids at an event in New York City, where the focus was really o Read more…

IBM Quantum Summit: Two New QPUs, Upgraded Qiskit, 10-year Roadmap and More

December 4, 2023

IBM kicks off its annual Quantum Summit today and will announce a broad range of advances including its much-anticipated 1121-qubit Condor QPU, a smaller 133-qu Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire