Parameterization of laboratories, has succeeded in miniaturizing and dramatically reducing the power consumption of the types of silicon-photonic components used today to transport bits around data centers via fiber optic cables. This equipment encodes data into multiple wavelengths of light from an infrared laser and sends the light through a fiber.
of Avicenna The chiplet couldn’t be more different: instead of infrared laser light, it uses ordinary light from a small screen in blue microLED. And instead of multiplexing all the optical data so it can travel on a single fiber, Avicena’s hardware sends the data in parallel through the separate lanes of a specialized optical cable.
Ayar has the weight of history on its side, offering customers similar technology to what they already use to send data over longer distances. But Avicena, the dark horse of this race, benefits from the continuous progress of the micro-display industry, which should
grow by 80% per year and reach US$123 billion by 2030fueled by a future full of virtual reality equipment and even augmented reality contact lenses.
“These companies are at opposite ends of the spectrum in terms of risk and innovation,” says
Vladimir Kozlovfounder and CEO of LightCounting, a telecommunications analytics company.
MicroLED vs Infrared Lasers
Avicena’s silicon chip, LightBundle, consists of an array of gallium nitride microLEDs, an array of equally sized photodetectors, and some I/O circuitry to support communication with the processor that it feeds with data. Twin 0.5 millimeter diameter optical cables connect the microLED array of one chiplet to the photodetectors of another and vice versa. These cables, similar to the imaging cables of some endoscopes, contain a bundle of fiber cores that align with the on-chip dies, giving each microLED its own light path.
Besides the existence of this type of cable, Avicena needed two other things to come together, explains
Bardia Pezeshki, CEO of the company. “The first, which I think was most surprising to anyone in the industry, is that LEDs could operate at 10 gigabits per second,” he says. “It’s mind-boggling” considering that the state of the art in visible light communication systems just five years ago was in the hundreds of megahertz ranges. But in 2021, Avicena researchers revealed a version of the microLED they dubbed Cavity-Enhanced Micro-Optical Emitters, or CROME. The devices are microLEDs that have been optimized for switching speed by minimizing capacitance and sacrificing some efficiency when converting electrons to light.
Gallium Nitride is not something that is typically integrated on silicon chips for computing, but thanks to advances in the microLED display industry, this is essentially a problem solved. In search of bright emissive displays for AR/VR and other things, tech giants such as Apple, Google and Meta have spent years finding ways to transfer already-built microscale LEDs to specific points on silicon and other surfaces. Today, “it’s done by the millions every day,” says Pezeshki. Avicene herself recently
bought the fab where he developed the CROMEs of his Silicon Valley neighbor Nanosys.
Computer makers will want solutions that will not only help in the next two to three years, but deliver reliable improvements for decades.
The second component was the photodetector. Silicon is not good at absorbing infrared light, so designers of silicon-photonics systems typically compensate by making photodetectors and other relatively large components. But because silicon readily absorbs blue light, Avicena’s system photodetectors only need a few tenths of a micrometer depth, allowing them to be easily integrated into the chip under the imaging fiber array. . Pezeshki credits Stanford David A. B. Miller by proving, more than ten years ago, that CMOS photodetectors detecting blue light were quick enough to get the job done.
The combination of imaging fiber, blue microLEDs and silicon photodetectors results in a system that, in prototypes, transmits “many” terabits per second, Pezeshki says. Just as important as the data rate is the low energy needed to move around a bit. “If you look at the target values for silicon photonics, they’re a few picojoules per bit, and those come from companies that are way ahead of us” in terms of commercialization, says Pezeshki. “We’ve already broken those records.” In one demo, the system moved data using about half a picojoule per bit. The startup’s first product, expected in 2023, will not reach the processor but will aim to connect servers within a data center rack. A chiplet for chip-to-chip optical links will follow “right on its heels”, says Pezeshki.
But there are limits to the ability of microLEDs to move data. Because LED light is incoherent, it suffers from scattering effects that limit it to around 10 meters. Lasers, on the other hand, are naturally good at going the distance; Ayar’s TeraPHY chips have a range of up to 2 kilometers, potentially disrupting the architecture of supercomputers and data centers even more than Avicena’s technology could. They could let computer makers completely rethink their architectures, allowing them to build “essentially a single computer chip, but building it rack-scale,” says
Charlie Wuischpard, CEO of Ayar. The company is ramping up production with partner GlobalFoundries and building prototypes with partners in 2023, though these are unlikely to be made public, he says.
Kozlov says he expects many more competitors to emerge. Computer makers will want solutions that “will not only be useful in the next two to three years, but provide reliable improvements for decades.” After all, the copper connections they seek to replace also continue to improve.
This article appears in the January 2023 print issue under the title “A Dark (Blue) Horse Emerges to Speed Up Computing”.
From articles on your site
Related articles on the web