Technical Roadmap

• 

  SRAM Compute-in-Memory

  Computing capabilities are integrated directly into SRAM cells, allowing operations to be performed where data is stored and reducing data movement. Houmo.AI continues to advance its proprietary CIM IP and chip architecture to support large-model inference across on-device and edge scenarios.

• 

  3D Compute-in-Memory

  To address the growing memory-access demands of large models, Houmo.AI has developed a new-generation 3D-CIM architecture. It combines multilayer 3D memory based on near-memory computing with SRAM-CIM, creating a “1+1>2” multiplier effect that efficiently addresses both computing requirements and memory bandwidth bottlenecks.