The massive market for bright and efficient displays has driven GaN microLED technology to enable megapixel displays with millions of devices transferred onto silicon backplanes. This same technology can be used to solve the data bottleneck currently chocking AI and ML applications, where processors need to move massive amounts of data between XPUs or memory. At Avicena we have shown that a modified version of these microLEDs can modulate at >10Gb/s, and with the parallelism of displays, transfer massive amounts of data at very high density using multicore or imaging fiber at sub picojoules per bit. This approach is well suited replace SERDES based techniques over copper for short distances and ease dataflow to memory or for HPC applications.

About the speaker:  Bardia received his Ph.D. EE in the 90's from Stanford working on quantum well optical modulators, one of many approaches considered to enable optical interconnects for chips. Since then, he developed a tunable laser at Santur that carried most of the world's long distance internet traffic for a number of years, and a multiwavelength optical transceiver technology at Kaiam that enabled 40G and 100G pluggable optics at the large datacenters. He has recently returned to his original aim of enabling optical chip to chip communication, but this time with GaN microLEDs - a device from the display world that he believes has magical powers to solve silicon's greatest problem of moving data efficiently.