CHIPSMORE: Compute-in-Interconnect and -Memory Chiplets for Multi-Mode Multi-Request LLM Inference Acceleration
2026-08-31 • Hardware Architecture
Hardware Architecture
AI summaryⓘ
The authors developed CHIPSMORE, a special computer chip designed to run large language models (LLMs) more efficiently. This chip combines different memory and processing types to handle multiple requests and long contexts without wasting energy or memory. Their design avoids copying large model weights when handling multiple tasks, saving a lot of power. Tests showed CHIPSMORE can work faster and use much less energy compared to a popular Nvidia chip during inference. Overall, the authors present a flexible and energy-saving solution for running language models on new hardware.
Large Language ModelsCompute-in-MemoryResistive RAMStatic RAMLow-Rank AdaptationInference AccelerationKey-Value MemoryMulti-request ExecutionEnergy EfficiencyBatch Processing
Authors
Yue Jiet Chong, Yimin Wang, Zhen Wu, Zixuan Wang, Wei Zhang, Xuanyao Fong
Abstract
Large language model (LLM) inference exhibits substantial variability across adaptation modes, context lengths, and request concurrency, creating challenges for maintaining high utilization, memory efficiency, and scalable performance on compute-in-memory (CIM) accelerators. This paper presents CHIPSMORE, a multi-mode and multi-request LLM inference accelerator that integrates compute-in-interconnect and CIM to support both base-mode and low-rank adaptation (LoRA) inference under diverse workloads. CHIPSMORE employs heterogeneous processing elements consisting of resistive RAM analog compute-in-memory (RRAM-ACIM) and static RAM digital compute-in-memory (SRAM-DCIM) interconnected through a programmable Inter-PE computational network (IPCN). A composable hierarchical key-value (KV) memory scheme dynamically allocates router scratchpad, SRAM-DCIM, and embedded DRAM (eDRAM) resources according to workload requirements, enabling scalable support for long-context and batched inference. Furthermore, a non-replicated multi-request execution pipeline exploits request-level parallelism without duplicating pretrained weights, while a state-aware resource reconfiguration mechanism selectively retains runtime states and power-gates inactive resources to improve energy efficiency. Evaluation using cycle-accurate hardware-software co-simulation demonstrates that CHIPSMORE effectively sustains high throughput across varying model sizes, context lengths, and batch sizes while maintaining favorable power scaling. Compared with Nvidia H100, CHIPSMORE achieves up to $2.38\times$ higher throughput and $27\times$ higher energy efficiency on Mistral-7B inference while eliminating weight replication for multi-request serving.