This paper presents BROCA, a low-power and low-latency mobile social agent system-on-chip with 3 key features. 1) ABTU with ACER supports decomposed dimension-aware compute data path for energy-efficient response generation. 2) ACBU with input bit width optimization for agent feedback vocoder compute power reduction; 3) LTMU for low-latency dialogue context RAG while retaining the conversational context. As a result, BROCA achieves 3.3× speedup, and 1.8× consumes less energy in mobile devices.

  - Agent compute engine (ACE) router and adaptive bit-truncate unit (ABTU)

  - Acoustic cluster bit-grouping unit (ACBU)

  - Long-term memory unit (LTMU) with and dual direct memory access (DMA)

  - ISSCC 2025