Light-enabled Microprocessor Holds Promise for Faster Computers

By John Russell

February 23, 2016

Combining electronics and photonics on semiconductor microchips to speed data transmission isn’t a new idea – the potential for better performance and power reduction are enticing. However thorny manufacturing issues have so far limited widespread use of this approach. That could change soon according to a recent report in Nature[i] and would have have broad implications extending even to efforts to achieve exascale computing, say the authors.

In the paper – “Single-chip microprocessor that communicates directly using light” – researchers from UC Berkeley, University of Colorado, and MIT report fabricating an electronic–photonic system on a single chip integrating over 70 million transistors and 850 photonic components that work together to provide logic, memory, and interconnect functions. Most significantly, they did it with standard CMOS manufacturing techniques.

Talking about the impact of the work, Miloš Popovic a co-author on the study from the University of Colorado told HPCwire, “This work is directly aimed at the energy problem in supercomputers.  It will enable reducing the communication energy by about an order of magnitude, and will make communication energy independent of distance of a link — up to 100’s of meters. So, it’s definitely part of the exascale computing story.”

The chip was fabricated using a commercial high-performance 45-nm complementary metal–oxide semiconductor (CMOS) silicon-on-insulator (SOI) process. The authors write: “No changes to the foundry process were necessary to accommodate photonics and all optical devices were designed to comply with the native process-manufacturing rules. This ‘zero-change’ integration enables high-performance transistors on the same chip as optics, reuse of all existing designs in the process, compatibility with electronics design tools, and manufacturing in an existing high-volume foundry.”

On-chip electro-optic transmitters and receivers enable both the microprocessor and the memory to communicate directly to off-chip components using light, without the need for separate chips or components to host the optical devices.

One advantage of light based communication, noted Popović, is that multiple parallel data streams encoded on different colors of light can be sent over one and the same medium – in this case, an optical wire waveguide on a chip, or an off-chip optical fiber of the same kind that as those that form the Internet backbone.

Close-up of light-enabled microprocessor showing optical circuits (left), memory (top) and 2 compute cores (right)
Close-up of light-enabled microprocessor showing optical circuits (left), memory (top) and 2 compute cores (right)

“Another advantage is that the infrared light that we use – and that also TV remotes use – has a physical wavelength shorter than 1 micron, about one hundredth of the thickness of a human hair,” he said. “This enables very dense packing of light communication ports on a chip, enabling huge total bandwidth.” The new chip has a bandwidth density of 300 gigabits per second per square millimeter, about 10 to 50 times greater than packaged electrical-only microprocessors currently on the market.

The big news is the relative ease of manufacture. “This “zero change” approach to integration enables complex electronic-photonic systems on chip to be designed today, in an advanced CMOS foundry. This means high yield, immediate transition to volume production, and the most advanced transistors of any photonic chip (and the largest number of them). These qualities should open up research into systems on chip in many applications including RF signal processing, radar/lidar applications, sensing and imaging, etc.

The authors note, “By showing that a microprocessor with photonic I/O is possible to build today, we’re illustrating the power of this approach.  Incidentally, while we expected photonic devices to not perform as well using this approach as using fabrication customized to photonics, it turns out that in a number of cases they perform better — leveraging the high resolution implant masks, controlled sub-100nm CMOS deep UV lithography, and rich set of material and mask levels available in CMOS.”

As described in the paper, the manufacturing process includes a crystalline-silicon layer that is patterned to form both the body of the electronic transistors and the core of the optical waveguides. A thin buried-oxide layer separates the crystalline-silicon layer from the silicon-handle wafer. Because the buried-oxide layer is <200 nm thick, light propagating in crystalline-silicon waveguides will evanescently leak into the silicon-handle wafer, resulting in high waveguide loss.

To control leakage, selective substrate removal was performed on the chips after electrical packaging to etch away the silicon handle under regions with optical devices. The silicon handle remains intact under the microprocessor and memory (which dissipate the most power) to allow a heat sink to be contacted, if necessary. The researchers report removal of substrate “has a negligible effect on the electronics and the processor is completely functional even with a fully removed substrate.” The full details are best gleaned from the paper itself.

Researchers built the photodetectors of Silicon-Germanium (SiGe), which is present in small amounts in advanced CMOS processing, and selected 1,180nm wavelength for the optical channels as silicon is transparent to that wavelength. No adverse effects were seen and propagation losses were 4.3 dB/cm. The electro-optic transmitter consists of an electro-optic modulator and its electronic driver. The modulator is a silicon micro-ring resonator with a diameter of 10 μm, coupled to a waveguide.

Electrical signals are encoded on light waves in this optical transmitter consisting of a spoked ring modulator, monitoring photodiode (left) and light access port (bottom), all built using the same manufacturing steps and alongside transistor circuits that control them (top)
Electrical signals are encoded on light waves in this optical transmitter consisting of a spoked ring modulator, monitoring photodiode (left) and light access port (bottom), all built using the same manufacturing steps and alongside transistor circuits that control them (top)

It was necessary to create a tuning mechanism report the authors: “As a resonant device, the modulator is highly sensitive to variations in the thickness of the crystalline-silicon layer within and across SOI wafers as well as to spatially and rapidly temporally varying thermal environments created by the electrical components on the chip. Both effects cause λ0 to deviate from the design value, necessitating tuning circuitry. We embedded a 400-Ω resistive microheater inside the ring to efficiently tune λ0 and added a monitoring photodetector weakly coupled to the modulator drop port. When light resonates in the modulator ring, a small fraction of it couples to and illuminates the photodetector.”

Sadasivan Shankar, a longtime senior Intel researcher in semiconductor manufacturing and currently a visiting lecturer in computational science and engineering at Harvard, called the work important. “As mentioned in the paper itself, this is the current strained transistor technology that has been available in the market. [Nevertheless] this is a significant milestone. An optical device for transmission to memory in principle can save energy and also increase the clock cycle,” said Shankar who was not associated with the work.

“The current paradigm in HPC is more moving towards taking computing to data. The integration of optics with electronics on the same chip could enable this without higher energy costs. However, it is not clear that the overall performance is competitive with the state­‐of­‐the-art 14 nm CMOS technology,” said Shankar.

Popovic noted the important next steps for the research include: “1) to demonstrate multi-wavelength (WDM) communication in a processor, and 2) to improve the photonic devices — both of which can be done — to really put to rest questions about the viability of the approach, and 3) to develop new system applications — that will in turn drive us to devise new device concepts within CMOS platforms.”

Challenges aside, the work is a significant step forward. Co-author Chen Sun of UC Berkley said, “At a high-level, our work could solve the interconnect problem of today’s chips inside computers; semiconductor technology has allowed us to do more and more compute on a chip, but has done little to help chips communicate with each other at a higher bandwidth. Furthermore, the amount of power chips spend on communicating with other chips is now >20% of the chip’s power budget.

“With this technology, we could improve chip communication bandwidth by more than an order of magnitude and at lower power. At a lower-level, we have demonstrated an alternative path towards making optical devices on microchips, one that could 100% rely on an existing microchip manufacturing process and be natively integrated with electronics. This is an alternative to how the field of silicon photonics makes devices today, which is typically with the development of a new manufacturing process which no ability to integrate transistors on-chip.”

[i] Single-chip microprocessor that communicates directly using light, Nature

528, 534–538 (24 December 2015) doi:10.1038/nature16454; http://www.nature.com/nature/journal/v528/n7583/full/nature16454.html

Top Photo: Glenn Asakawa
Second Photo: Milos Popovic
Third Photo: Mark Wade

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 Training 4.0 – Nvidia Still King; Power and LLM Fine Tuning Added

June 12, 2024

There are really two stories packaged in the most recent MLPerf  Training 4.0 results, released today. The first, of course, is the results. Nvidia (currently king of accelerated computing) wins again, sweeping all nine Read more…

Highlights from GlobusWorld 2024: The Conference for Reimagining Research IT

June 11, 2024

The Globus user conference, now in its 22nd year, brought together over 180 researchers, system administrators, developers, and IT leaders from 55 top research computing centers, national labs, federal agencies, and univ Read more…

Nvidia Shipped 3.76 Million Data-center GPUs in 2023, According to Study

June 10, 2024

Nvidia had an explosive 2023 in data-center GPU shipments, which totaled roughly 3.76 million units, according to a study conducted by semiconductor analyst firm TechInsights. Nvidia's GPU shipments in 2023 grew by more Read more…

Weekly Wire Roundup: June 2-June 7, 2024

June 8, 2024

Computex (and Jensen Huang) gave us an extra day of news this week, compensating for last week's shorter, holiday-driven news cycle. On Sunday ahead of the official start of Computex, Nvidia's CEO Jensen Huang deliver Read more…

ASC24 Expert Perspective: Dongarra, Hoefler, Yong Lin

June 7, 2024

One of the great things about being at an ASC (Asia Supercomputer Community) cluster competition is getting the chance to interview various industry experts and learning more about the various challenges the students are Read more…

HPC and Climate: Coastal Hurricanes Around the World Are Intensifying Faster

June 6, 2024

Hurricanes are among the world's most destructive natural hazards. Their environment shapes their ability to deliver damage; conditions like warm ocean waters, guiding winds, and atmospheric moisture can all dictate stor Read more…

MLPerf Training 4.0 – Nvidia Still King; Power and LLM Fine Tuning Added

June 12, 2024

There are really two stories packaged in the most recent MLPerf  Training 4.0 results, released today. The first, of course, is the results. Nvidia (currently Read more…

Highlights from GlobusWorld 2024: The Conference for Reimagining Research IT

June 11, 2024

The Globus user conference, now in its 22nd year, brought together over 180 researchers, system administrators, developers, and IT leaders from 55 top research Read more…

Nvidia Shipped 3.76 Million Data-center GPUs in 2023, According to Study

June 10, 2024

Nvidia had an explosive 2023 in data-center GPU shipments, which totaled roughly 3.76 million units, according to a study conducted by semiconductor analyst fir Read more…

ASC24 Expert Perspective: Dongarra, Hoefler, Yong Lin

June 7, 2024

One of the great things about being at an ASC (Asia Supercomputer Community) cluster competition is getting the chance to interview various industry experts and Read more…

HPC and Climate: Coastal Hurricanes Around the World Are Intensifying Faster

June 6, 2024

Hurricanes are among the world's most destructive natural hazards. Their environment shapes their ability to deliver damage; conditions like warm ocean waters, Read more…

ASC24: The Battle, The Apps, and The Competitors

June 5, 2024

The ASC24 (Asia Supercomputer Community) Student Cluster Competition was one for the ages. More than 350 university teams worked for months in the preliminary competition to earn one of the 25 final competition slots. The winning teams... Read more…

Computex 2024: Nvidia, AMD Push GPUs; Intel Revs Up x86 Power Efficiency

June 5, 2024

"The days of millions of GPU data centers are coming," said Nvidia CEO Jensen Huang during a keynote at Computex. Huang's predictions are becoming bolder and bo Read more…

Using AI and Robots to Advance Science

June 4, 2024

Even though we invented it, humans can be pretty bad at science. We need to eat and sleep, we sometimes let our emotions regulate our behavior, and our bodies a Read more…

Atos Outlines Plans to Get Acquired, and a Path Forward

May 21, 2024

Atos – via its subsidiary Eviden – is the second major supercomputer maker outside of HPE, while others have largely dropped out. The lack of integrators and Atos' financial turmoil have the HPC market worried. If Atos goes under, HPE will be the only major option for building large-scale systems. Read more…

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…

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…

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…

Everyone Except Nvidia Forms Ultra Accelerator Link (UALink) Consortium

May 30, 2024

Consider the GPU. An island of SIMD greatness that makes light work of matrix math. Originally designed to rapidly paint dots on a computer monitor, it was then 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…

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…

Some Reasons Why Aurora Didn’t Take First Place in the Top500 List

May 15, 2024

The makers of the Aurora supercomputer, which is housed at the Argonne National Laboratory, gave some reasons why the system didn't make the top spot on the Top Read more…

Leading Solution Providers

Contributors

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…

The NASA Black Hole Plunge

May 7, 2024

We have all thought about it. No one has done it, but now, thanks to HPC, we see what it looks like. Hold on to your feet because NASA has released videos of wh Read more…

GenAI Having Major Impact on Data Culture, Survey Says

February 21, 2024

While 2023 was the year of GenAI, the adoption rates for GenAI did not match expectations. Most organizations are continuing to invest in GenAI but are yet to Read more…

Google Announces Sixth-generation AI Chip, a TPU Called Trillium

May 17, 2024

On Tuesday May 14th, Google announced its sixth-generation TPU (tensor processing unit) called Trillium.  The chip, essentially a TPU v6, is the company's l 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…

Intel’s Next-gen Falcon Shores Coming Out in Late 2025 

April 30, 2024

It's a long wait for customers hanging on for Intel's next-generation GPU, Falcon Shores, which will be released in late 2025.  "Then we have a rich, a very Read more…

Intel Plans Falcon Shores 2 GPU Supercomputing Chip for 2026  

August 8, 2023

Intel is planning to onboard a new version of the Falcon Shores chip in 2026, which is code-named Falcon Shores 2. The new product was announced by CEO Pat Gel 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…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire