Brainscaleintelligencewithatomicscaletechnology

With only a 28 nm process, we enable:

• 330GB @ 2.1 PB/s DRAM on die that obsoletes HBM and interposers
• 14,438 tokens/s80B-parameter decode at FP8 on a single die
• 174× tokens/W25.8 mJ per token vs ~4.5 J for HBM4 GPUs
• Space-gradeIntrinsically rad-hard — cosmic rays leave no lattice damage