Home / Technology / AI Chip Leaps Forward: First 12nm RRAM SoC Taped Out
AI Chip Leaps Forward: First 12nm RRAM SoC Taped Out
27 Oct
Summary
- TetraMem taped out the world's first 12nm RRAM AI chip, MLX300.
- Synopsys Cloud EDA accelerated distributed development across US and Singapore.
- MLX300 supports 64 million parameters without external DRAM.
TetraMem, a Silicon Valley-based company specializing in analog in-memory computing (A-IMC), announced the successful tapeout of its MLX300 chip. This groundbreaking SoC is the world's first 12nm device to utilize multi-level RRAM technology for AI applications. The development was significantly accelerated by leveraging Synopsys Cloud FlexEDA solutions, enabling efficient global collaboration.
The MLX300 tapeout follows TetraMem's earlier success with the MLX200/MLX200M on a 22nm process, with both achieving full-mask tapeouts in 2026. The use of Synopsys' cloud-enabled EDA tools dramatically improved design productivity, reducing environment setup time from weeks to minutes and achieving faster simulation speeds compared to on-premises methods.
This advanced RRAM-based architecture aims to deliver superior performance-per-watt for AI accelerators. The MLX300 is designed as a versatile chip for audio, vision, and small language models, capable of handling up to 64 million INT4 parameters internally. It targets an energy efficiency of over 50-80 TOPS/W, highlighting the potential of A-IMC in next-generation AI inference.
