#semianalysis

Lightmatter made a big splash a couple of years ago by presenting an AI accelerator, Mars, that rethought the paradigm of how electronics compute and move data. They presented a processor that used photonics for compute. This chip promised multiple orders of magnitude improvement in latency, bandwidth, and power. This ultimately was unsuccessful due to the rigidity of the software and compute structure involved with the chip. Lightmatter has created a 2nd generation of this AI compute produc...

Lightmatter made a big splash a couple of years ago by presenting an AI accelerator, Mars, that rethought the paradigm of how electronics compute and move data. They presented a processor that used photonics for compute. This chip promised multiple orders of magnitude improvement in latency, bandwidth, and power. This ultimately was unsuccessful due to the rigidity of the software and compute structure involved with the chip. Lightmatter has created a 2nd generation of this AI compute produc...

Lightmatter made a big splash a couple of years ago by presenting an AI accelerator, Mars, that rethought the paradigm of how electronics compute and move data. They presented a processor that used photonics for compute. This chip promised multiple orders of magnitude improvement in latency, bandwidth, and power. This ultimately was unsuccessful due to the rigidity of the software and compute structure involved with the chip. Lightmatter has created a 2nd generation of this AI compute produc...

Lightmatter made a big splash a couple of years ago by presenting an AI accelerator, Mars, that rethought the paradigm of how electronics compute and move data. They presented a processor that used photonics for compute. This chip promised multiple orders of magnitude improvement in latency, bandwidth, and power. This ultimately was unsuccessful due to the rigidity of the software and compute structure involved with the chip. Lightmatter has created a 2nd generation of this AI compute produc...

Lightmatter made a big splash a couple of years ago by presenting an AI accelerator, Mars, that rethought the paradigm of how electronics compute and move data. They presented a processor that used photonics for compute. This chip promised multiple orders of magnitude improvement in latency, bandwidth, and power. This ultimately was unsuccessful due to the rigidity of the software and compute structure involved with the chip. Lightmatter has created a 2nd generation of this AI compute produc...

Lightmatter made a big splash a couple of years ago by presenting an AI accelerator, Mars, that rethought the paradigm of how electronics compute and move data. They presented a processor that used photonics for compute. This chip promised multiple orders of magnitude improvement in latency, bandwidth, and power. This ultimately was unsuccessful due to the rigidity of the software and compute structure involved with the chip. Lightmatter has created a 2nd generation of this AI compute produc...

NAND is a market with fierce competition and a relentless march of progress. The number of bits of NAND manufactured and shipped has grown at a pace of 30% to 35% every year, doubling every 2 to 3 years. The initial take would be that this requires a large amount of capital dedicated to new equipment, but the NAND industry has only spent from $15B to $20B on wafer fabrication equipment every year from 2017 to 2022 despite exponential growth. The cost to produce NAND has fallen rapidly. Simil...

NAND is a market with fierce competition and a relentless march of progress. The number of bits of NAND manufactured and shipped has grown at a pace of 30% to 35% every year, doubling every 2 to 3 years. The initial take would be that this requires a large amount of capital dedicated to new equipment, but the NAND industry has only spent from $15B to $20B on wafer fabrication equipment every year from 2017 to 2022 despite exponential growth. The cost to produce NAND has fallen rapidly. Simil...