# HALO: Memory-Centric Heterogeneous Accelerator with 2.5D Integration for Low-Batch LLM Inference

arXiv: https://arxiv.org/abs/2510.02675 · HTML: https://mlsanity.com/paper/2510.02675/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 24 of its 37 references that are in the map.
- First posted on arXiv 3 Oct 2025.

## Builds on (most cited first, 24 of 24)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Scaling Autoregressive Models for Content-Rich Text-to-Image Generation](https://mlsanity.com/paper/2206.10789/) (2022)
- [VideoChat: Chat-Centric Video Understanding](https://mlsanity.com/paper/2305.06355/) (2023)
- [QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMs](https://mlsanity.com/paper/2404.00456/) (2024)
- [A Survey on Large Language Models for Recommendation](https://mlsanity.com/paper/2305.19860/) (2023)
- [LLM in a flash: Efficient Large Language Model Inference with Limited Memory](https://mlsanity.com/paper/2312.11514/) (2023)
- [AnnoLLM: Making Large Language Models to Be Better Crowdsourced Annotators](https://mlsanity.com/paper/2303.16854/) (2023)
- [GEAR: An Efficient KV Cache Compression Recipe for Near-Lossless Generative Inference of LLM](https://mlsanity.com/paper/2403.05527/) (2024)
- [Fluctuation-based Adaptive Structured Pruning for Large Language Models](https://mlsanity.com/paper/2312.11983/) (2023)
- [Large Language Models Meet NLP: A Survey](https://mlsanity.com/paper/2405.12819/) (2024)
- [Enhancing Recommender Systems with Large Language Model Reasoning Graphs](https://mlsanity.com/paper/2308.10835/) (2023)
- [NeuPIMs: NPU-PIM Heterogeneous Acceleration for Batched LLM Inferencing](https://mlsanity.com/paper/2403.00579/) (2024)
- [X-Former: In-Memory Acceleration of Transformers](https://mlsanity.com/paper/2303.07470/) (2023)
- [PIM Is All You Need: A CXL-Enabled GPU-Free System for Large Language Model Inference](https://mlsanity.com/paper/2502.07578/) (2025)
- [Faster and Lighter LLMs: A Survey on Current Challenges and Way Forward](https://mlsanity.com/paper/2402.01799/) (2024)
- [PrefillOnly: An Inference Engine for Prefill-only Workloads in Large Language Model Applications](https://mlsanity.com/paper/2505.07203/) (2025)
- [Depression Detection on Social Media with Large Language Models](https://mlsanity.com/paper/2403.10750/) (2024)
- [Beyond Classification: Financial Reasoning in State-of-the-Art Language Models](https://mlsanity.com/paper/2305.01505/) (2023)
- [Puzzle: Distillation-Based NAS for Inference-Optimized LLMs](https://mlsanity.com/paper/2411.19146/) (2024)
- [LPZero: Language Model Zero-cost Proxy Search from Zero](https://mlsanity.com/paper/2410.04808/) (2024)
- [COMET: A Framework for Modeling Compound Operation Dataflows with Explicit Collectives](https://mlsanity.com/paper/2509.00599/) (2025)

## Abstract (arXiv)

The rapid adoption of Large Language Models (LLMs) has driven a growing demand for efficient inference, particularly in latency-sensitive applications such as chatbots and personalized assistants. Unlike traditional deep neural networks, LLM inference proceeds in two distinct phases: the prefill phase, which processes the full input sequence in parallel, and the decode phase, which generates tokens sequentially. These phases exhibit highly diverse compute and memory requirements, which makes accelerator design particularly challenging. Prior works have primarily been optimized for high-batch inference or evaluated only short input context lengths, leaving the low-batch and long context regime, which is critical for interactive applications, largely underexplored. We propose HALO, a heterogeneous memory centric accelerator designed for these unique challenges of prefill and decode phases in low-batch LLM inference. HALO integrates HBM based Compute-in-DRAM (CiD) with an on-chip analog Compute-in-Memory (CiM), co-packaged using 2.5D integration. To further improve the hardware utilization, we introduce a phase-aware mapping strategy that adapts to the distinct demands of the prefill and decode phases. Compute bound operations in the prefill phase are mapped to CiM to exploit its high throughput matrix multiplication capability, while memory-bound operations in the decode phase are executed on CiD to benefit from reduced data movement within DRAM. Additionally, we present an analysis of the performance tradeoffs of LLMs under two architectural extremes: a fully CiD and a fully on-chip analog CiM design to highlight the need for a heterogeneous design. We evaluate HALO on LLaMA-2 7B and Qwen3 8B models. Our experimental results show that LLMs mapped to HALO achieve up to 18x geometric mean speedup over AttAcc, an attention-optimized mapping and 2.5x over CENT, a fully CiD based mapping.