The transition from electronics to photonics is expected to be the next digital frontier and norm of the future. From networks to 6G and chip-to-chip communication, photonics is poised as the solution for more powerful, energy-efficient, high-speed data transfers.
Intel, Sony, NTT, and countless other companies and startups are developing new photonics technologies that reimagine how we use chips and networks.
Computing today still has a couple of bottlenecks ahead of it — one being that electrons have a maximum speed at which they can send data back and forth.
But if the clue is not in the name, photonics uses light — and nothing beats light.
Techopedia spoke to Bardia Pezeshki, PhD, CEO and board member of Avicena, a high-bandwidth semiconductor company based in Sunnyvale, California, and one of the companies exploring chip-to-chip communication and connectivity with ultra-fast µLEDs, or micro-LEDs.
Key Takeaways
- By harnessing the power of light, photonics offers significant advantages over traditional electronics in terms of speed, efficiency, and data transfer capacity.
- Photonics is crucial for overcoming the limitations of current computing architectures, particularly in areas like AI and high-performance computing.
- To fully realize the potential of photonics, collaboration between tech giants, startups, and research institutions is vital for developing standards and overcoming technical hurdles.
- The widespread adoption of photonics will revolutionize various industries, from data centers to healthcare, as it drives unprecedented advancements in technology and performance.
Compute Challenges: Energy, Processing Power, and Low Latency
As artificial intelligence (AI) and machine learning (ML) pick up speed and momentum, with companies like NVIDIA releasing more powerful chips and others like Intel and Huawei going all in to compete, a big problem remains unresolved:
AI and ML are processing more data than ever, can cause high latency, and are driving data center energy consumption to an all-time high.
Different studies predict a significant rise in energy consumption driven by AI. For example, Goldman Sachs found that the tech is expected to increase data center power demand by 160%.
Pezeshki from Avicena explained that for decades people have wanted to connect ICs with photonics, saying:
“The reason is that photons are much better at moving information than electrons. They don’t interact with each other and there is no electrical resistance and capacitance.”
“Chips That Talk With Light’
The potential of photonics, with Intel having invested billions of dollars in silicon photonics and optical technologies over the last few decades, has now attracted the attention of most big tech players, from Lockheed Martin, IAG Capital Partners, Hewlett Packard Enterprises, and NVIDIA, as mentioned.
However, photonic technology is still developing and faces many obstacles. Pezeshki breaks it down for us.
“The problem of moving data on and off chips is the biggest bottleneck in electronics. Almost all advanced chips are (Input/Output) I/O limited, and this limitation is most acute in AI-type configurations where GPUs have to work in parallel through a network of switches.”
Pezeshki explained that the key problem for photonics chips is that silicon is fundamentally not a good optical material and not compatible with photonic standard lasers.
“It is really too expensive and complicated to connect up chips that are only a meter or so away from each other using standard fiber optics technology.”
Avicena’s solution? To use LEDs and display technology to make ‘chips talk with light’.
“We basically put a small microLED display and a camera on chips and let the chips ‘look’ at each other to exchange data through something like imaging fibers.”
To work with chips, Avicena’s microLED displays are simpler than normal displays —
1000 pixels of a single color versus millions of RGB. But with simplicity, Avicena gained high-speed performance — many Gb/s compared to the frame rate of a normal display.
“The display and camera technologies that we use are compatible with pretty much all (Complementary Metal-Oxide-Semiconductor) CMOS (the foundation for most modern electronic devices)”
Long Distance Photonics Vs. Short Distance Communications
NTT has already proven that photonics networks and the transfer of high speed, low latency, and mass volumes of data are possible.
NTT’s Research and Development department considers photonics a natural evolution that will have endless applications from remote advanced robotic real-time medical procedures and surgeries to digital twins, smart cars and smart cities, and AI data over networks.
But using photonics to transfer data under long-distance architectures is one thing, and making chips ‘talk’ in very short distances within the hardware is another. Pezeshki believes not only that it is possible but assertively claims his company is working to achieve this milestone.
“Our goal is to make that decade-long dream of photonics interconnects between chips come true.”
How is Avicena’s tech different from others who went down the same road, tried, but failed? Compared to traditional methods, their LED-based approach is much more energy efficient than copper or lasers. It is also much more parallel — think super-wide data busses that interface directly with how information moves natively on chips.
“There is no need to serialize the data to very high speeds per lane that needs even more electrical power and delays the movement of data,” Pezeshki said.
As AI clusters face data challenges and become more hungry, Pezeshki said optical interfaces on memory chips and within the GPUs allow each processor to access more memory and for many processors to share a big pool of memory.
Use Cases and Applications: Photonics in Real Life
Photonic chip-to-chip communication may represent a giant shift in computing architecture.
By harnessing the speed and efficiency of light, this technology overcomes the fundamental limitations of traditional electrical interconnects. Instead of relying on copper wires, photonic chips utilize light to transmit data between processors, memory units, and other components.
Pezeshki spoke about why connections can be as important as processors.
“Big picture — computing power comes not only from fast processors and memory but also from the density of connections between the processors.”
“When one moves to photonics, especially a parallel configuration with lots of lanes, and information is flowing vertically off chips, you solve the problem head-on.”
Transforming global hardware to support photonics is a monumental task. Pezeshki explained how Avicena is working to make this transition more seamless.
“It is a big deal for Nvidia or Intel to spin a new processor with an intimate optical interface,” Pezeshki said. “It is important to provide a step-by-step roadmap for this transition.”
“We are making standard-compliant optical cables that can replace electric ones and board-mounted optics that don’t require a redesign of the GPU.”
As these optical solutions prove themselves with extensive field use, chip designers will get more comfortable getting the optics closer to the IC. Ultimately, of course, our technology can be incorporated into the processor and memory chips themselves and eliminate the interface packaging and ICs.
The Bottom Line
The shift from electronics to photonics is poised to revolutionize computing architecture by taking what some may see as an unexpected shift. Innovation that comes in the shape of software and code is widely different than that which arrives at the hardware level.
Making the shift from electronics and copper wires to direct chip-to-chip optical cables is challenging. Standardization across the industry, supply chain and logistics issues, and a full embrace from the industry are all factors needed for photonics to happen.
However, companies like Avicena, creating compliant, innovative solutions, are not alone. From Intel to NTT and all of big tech, photonics is on their agenda. The reason why is simple.
To journey further into the future of technology, the world needs faster and more efficient processing and transfers but also denser connections to allow for simultaneous communications. Photonics and light provide all that and more.