Ayar Labs to Accelerate Development and Application of Optical Interconnects in Artificial Intelligence/Machine Learning Architectures with NVIDIA

SANTA CLARA, Calif. May 25, 2022  Ayar Labs, the lea짯der in chip-to-chip opti짯cal con짯nec짯ti짯vi짯ty, is deve짯lo짯ping with NVIDIA ground짯brea짯king arti짯fi짯ci짯al intel짯li짯gence (AI) infra짯struc짯tu짯re based on opti짯cal I/O tech짯no짯lo짯gy to meet future demands of AI and high per짯for짯mance com짯pu짯ting (HPC) workloads. The col짯la짯bo짯ra짯ti짯on will focus on inte짯gra짯ting Ayar Labs tech짯no짯lo짯gy to deve짯lop sca짯le-out archi짯tec짯tures enab짯led by high-band짯width, low-laten짯cy and ultra-low-power opti짯cal-based inter짯con짯nects for future NVIDIA pro짯ducts. Tog짯e짯ther, the com짯pa짯nies plan to acce짯le짯ra짯te the deve짯lo짯p짯ment and adop짯ti짯on of opti짯cal I/O tech짯no짯lo짯gy to sup짯port the explo짯si짯ve growth of AI and machi짯ne lear짯ning (ML) appli짯ca짯ti짯ons and data volumes.

Mee짯ting Future Per짯for짯mance and Power Requi짯re짯ments with Opti짯cal I/O

Opti짯cal I/O uni짯que짯ly chan짯ges the per짯for짯mance and power tra짯jec짯to짯ries of sys짯tem designs by enab짯ling com짯pu짯te, memo짯ry and net짯wor짯king ASICs to com짯mu짯ni짯ca짯te with dra짯ma짯ti짯cal짯ly increased band짯width, at lower laten짯cy, over lon짯ger distances and at a frac짯tion of the power of exis짯ting elec짯tri짯cal I/O solu짯ti짯ons. The tech짯no짯lo짯gy is also foun짯da짯tio짯nal to enab짯ling emer짯ging hete짯ro짯ge짯neous com짯pu짯te sys짯tems, disaggregated/pooled designs, and uni짯fied memo짯ry archi짯tec짯tures that are cri짯ti짯cal to acce짯le짯ra짯ting future data cen짯ter innovation.

Today셲 sta짯te-of-the-art AI/ML trai짯ning archi짯tec짯tures are limi짯t짯ed by cur짯rent cop짯per-based com짯pu짯te-to-com짯pu짯te inter짯con짯nects to build sca짯le-out sys짯tems for tomorrow셲 requi짯re짯ments, said Charles Wuisch짯pard, CEO of Ayar Labs. 쏰ur work with NVIDIA to deve짯lop next-gene짯ra짯ti짯on solu짯ti짯ons based on opti짯cal I/O pro짯vi짯des the foun짯da짯ti짯on for the next leap in AI capa짯bi짯li짯ties to address the world셲 most sophisti짯ca짯ted problems.

Deli짯ve짯ring the 쁍ext Million멭 Spee짯dup for AI with Opti짯cal Interconnect

Over the past deca짯de, NVI짯DIA-acce짯le짯ra짯ted com짯pu짯ting has deli짯ver짯ed a million멭 spee짯dup in AI, said Rob Ober, Chief Plat짯form Archi짯tect for Data Cen짯ter Pro짯ducts at NVIDIA. 쏷he next million멭 will requi짯re new, advan짯ced tech짯no짯lo짯gies like opti짯cal I/O to sup짯port the band짯width, power and sca짯le requi짯re짯ments of future AI and ML workloads and sys짯tem architectures.

As AI model sizes con짯ti짯nue to grow, by 2023 NVIDIA belie짯ves that models will have 100 tril짯li짯on or more con짯nec짯tions  a 600X increase from 2021 excee짯ding the tech짯ni짯cal capa짯bi짯li짯ties of exis짯ting plat짯forms. Tra짯di짯tio짯nal elec짯tri짯cal-based inter짯con짯nects will reach their band짯width limits, dri짯ving lower appli짯ca짯ti짯on per짯for짯mance, hig짯her laten짯cy and increased power con짯sump짯ti짯on. New inter짯con짯nect solu짯ti짯ons and sys짯tem archi짯tec짯tures are nee짯ded to address the sca짯le, per짯for짯mance and power demands of the next gene짯ra짯ti짯on of AI. Ayar Labs col짯la짯bo짯ra짯ti짯on with NVIDIA is focu짯sed on addres짯sing the짯se future chal짯lenges by deve짯lo짯ping next-gene짯ra짯ti짯on archi짯tec짯tures with opti짯cal I/O.

To learn more about Ayar Labs chip-to-chip opti짯cal tech짯no짯lo짯gy, plea짯se visit: https://ayarlabs.com/

About Ayar Labs

Ayar Labs is dis짯rupt짯ing the tra짯di짯tio짯nal per짯for짯mance, cost, and effi짯ci짯en짯cy cur짯ves of the semi짯con짯duc짯tor and com짯pu짯ting indus짯tries by dri짯ving a 1000x impro짯ve짯ment in inter짯con짯nect band짯width den짯si짯ty at 10x lower power.  Ayar Labs paten짯ted approach uses indus짯try stan짯dard cost-effec짯ti짯ve sili짯con pro짯ces짯sing tech짯ni짯ques to deve짯lop high speed, high den짯si짯ty, low power opti짯cal-based inter짯con짯nect 쐁hip짯lets and lasers to replace tra짯di짯tio짯nal elec짯tri짯cal based I/O. The com짯pa짯ny was foun짯ded in 2015 and is fun짯ded by a num짯ber of dome짯stic and inter짯na짯tio짯nal ven짯ture capi짯tal firms as well as stra짯te짯gic inves짯tors.  For more infor짯ma짯ti짯on, visit www.ayarlabs.com.