Industry has shrunk transistors down to a few atoms wide. Transistor-based silicon chips have increased computing power exponentially over the past few decades, but shrinking transistors further will be challenging, not just because shrinking becomes harder. , the signal also has the problem of loss. Many large-scale machine learning algorithms can be performed on tens of thousands of chips, but the speed at which data can be transferred between chips or servers using current current transfer methods is a bottleneck.

While humans have used light to transmit data through fiber-optic cables, including undersea cables, for decades, getting to the chip level is difficult because the devices used to generate or control light are not yet easily accessible. shrunk to the size of a transistor. Moore's Law, which doubles the density of transistors on a chip and reduces costs every two years, is slowing, prompting the industry to seek new solutions to handle the increasingly heavy demands of AI computing.

A few years ago, the technology of computers using "light" instead of "electricity" for computing processing and data transmission was only regarded as a research project. Now that the technology is accelerating, startups that can solve the engineering challenges associated with applying photonics on a chip are getting huge funding.

Silicon photonics startups raised more than $750 million last year, doubling from 2020 and well above the $18 million raised in 2016, according to PitchBook data. Brendan Burke, senior emerging technology analyst at PitchBook, predicts that by 2025, silicon photonics will become common hardware in data centers, and the market is estimated to reach $3 billion by then.

Ayar Labs, a startup developing silicon photonics technology, announced on the 26th that it has raised $130 million from investors including Nvidia and Intel. Bill Dally, chief scientist and senior vice president of research at NVIDIA, said: "Optical connectivity is important for scaling accelerated computing clusters to meet the rapidly growing demands of AI and HPC workloads. AyarLabs has unique optical I/O technology that enables scaling of next-generation silicon-based Photonics’ AI architecture needs.” said Ayar Labs CEO, “This financing allows us to start this year fully in accordance with industry standards to verify the quality and reliability of the solution as well as scale production.”

Charles Wuischpard, CEO of Ayar Labs, said: "The overall financing scale is significantly larger than our original goal, which underscores the market opportunity for optical I/O and Ayar Labs' leadership in silicon photonics-based interconnect solutions."

Ayar Labs also said: "The role of AI in data centers is becoming more and more important. The challenge of moving data and the energy consumption of moving data is a big problem." Ayar Labs revealed that the company has shipped the first batch of products in batches , expects to ship thousands of optical interconnect chips by the end of this year.

In addition to linking transistor chips, startups that use electrophotonics to build quantum computers, supercomputers and chips for self-driving cars have raised big bucks. For example, quantum computing company PsiQuantum has raised $665 million so far, although the prospect of quantum computers changing the world is still years away; and Lightmatter, a photonic chip research company, has raised $113 million and will launch chips later this year and work with customers Tests; Luminous Computing, which builds AI supercomputers, raised $115 million. The company also builds AI supercomputers using silicon photonics technology and has received funding from Microsoft founder Bill Gates.

In fact, not only start-up companies develop silicon photonics technology, but many semiconductor giants and some countries have also begun to deploy the photonics field.

Intel

In addition to the aforementioned investment in Ayar Labs, in December 2021 Intel Labs announced the establishment of a new research center aimed at driving faster and more efficient computing interfaces by replacing light with electricity . The news quickly attracted a lot of attention. It is reported that the goal of the Intel Research Center for Integrated Photonics for Data Center Interconnects (Intel Research Center for Integrated Photonics for Data Center Interconnects) is to solve the above bottlenecks by applying the same logic to the computing interface, especially for computing platforms to achieve breakthroughs .

More specifically, the research center, comprising researchers from leading academic institutions around the world, is developing optical input/output (I/O) technologies including silicon optoelectronics, CMOS circuit and link structures, package integration, and fiber coupling . Intel believes that optical I/O is expected to significantly outperform electrical in key performance metrics such as reach, bandwidth density, power consumption and latency, and this potential cannot be ignored.

James Jaussi, director of Intel Labs PHY Research Laboratory, pointed out that despite this potential, it is still difficult for optical interfaces to eliminate the previous copper interconnection in a short time, and related technical progress also needs to be gradual. As the masses transition from electronic and optical technologies, it needs a technology that can continue to connect and benefit generations, which will also be the focus of this research center.

GF

In March, GF announced its next-generation widely disruptive silicon photonics platform, GF Fotonix. GF is also partnering with Cisco Systems to customize silicon photonics solutions for DCN and DCI applications.

GF Fotonix is ??a monolithic platform and the first in the industry to integrate its differentiated 300mm photonics features and 300GHz-class RF-CMOS on a silicon chip, delivering best-in-class performance at scale. GF Fotonix integrates complex processes that were previously distributed across multiple chips onto a single chip by combining photonic systems, radio frequency (RF) components and high-performance complementary metal-oxide-semiconductor (CMOS) logic on a single silicon chip.

GF is the only pure-play foundry with a 300mm monolithic silicon photonics solution demonstrating the industry's highest single-fiber data rate (0.5Tbps/fiber). This enables 1.6-3.2Tbps optical chips to provide faster and more efficient data transmission, as well as more efficient signal integrity. In addition, the system error rate is increased by up to 10,000 times, enabling the next generation of artificial intelligence.

GF Fotonix enables the highest level of integration on photonic integrated circuits (PICs) so customers can integrate more product features and simplify their bill of materials (BOM). End customers can achieve higher performance by adding capacity and capability. The new solution also supports innovative packaging solutions such as passive connectivity for large fiber arrays, 2.5D packaging and on-chip lasers.

GF Fotonix solutions will be produced at the company's state-of-the-art manufacturing facility in Malta, New York, and PDK 1.0 will be available in April 2022. EDA partners Ansys, CadenceDesign Systems and Synopsys provide design tools and flows to support GF's customers and their solutions. GF provides customers with reference design kits, MPW, test, pre- and post-process, and semiconductor manufacturing services to help customers get to market faster.

Canada

In March, the Canadian government announced it was investing C$240 million ($187 million) to boost the country's semiconductor and photonics sectors, with a view to strengthening its role in the global market. C$90 million will be used to upgrade critical equipment at the Canadian Photonics Manufacturing Centre. Under the purview of the Government of Canada's National Research Council, which provides engineering and manufacturing services to academia and photonics businesses, it is expected to further enhance Canada's reputation as a "world leader in photonics."

The $150 million for the semiconductor fund is just a fraction of the $52 billion in semiconductor industry funding that the U.S. is awaiting final approval from Congress. But the Canadian government says the amount represents the country's "initial commitment" to its semiconductor industry, which includes more than 100 domestic and international companies that research and develop microchips, more than 30 applied research laboratories and five commercial facilities .