CQ-CiM: Hardware-Aware Embedding Shaping for Robust CiM-Based Retrieval Conference: Make sure to enter the correct conference title from your rights confirmation emai; July 26–29, 2026; Longbeach, CA
Xinzhao Li* 1 , Alptekin Vardar* 2 , Franz Müller 2 , Navya Goli 3 , Umamaheswara Rao Tida 3 , Kai Ni 4 , X. Sharon Hu 4 , Thomas Kämpfe 2 , Ruiyang Qin 1 1 Villanova University, 2 Fraunhofer IPMS, 3 NDSU, 4 University of Notre Dame · 2026
Read CQ-CiM: Hardware-Aware Embedding Shaping for Robust CiM-Based Retrieval Conference: Make sure to enter the correct conference title from your rights confirmation emai; July 26–29, 2026; Longbeach, CA by Xinzhao Li* 1 , Alptekin Vardar* 2 , Franz Müller 2 , Navya Goli 3 , Umamaheswara Rao Tida 3 , Kai Ni 4 , X. Sharon Hu 4 , Thomas Kämpfe 2 , Ruiyang Qin 1 1 Villanova University, 2 Fraunhofer IPMS, 3 NDSU, 4 University of Notre Dame (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-20083.txt.
Description
Retrieval-augmented generation on edge devices stalls on moving massive data through traditional architectures. Compute-in-memory promises relief by searching vectors directly inside crossbar arrays. A 2026 arXiv paper by Xinzhao Li and eight co-authors tackles the representation gap that blocks that promise.
High-precision, high-dimension embeddings do not fit low-precision, low-dimension memory arrays. Each memory design imposes its own cell and array constraints. Existing data shaping handles dimension and precision apart, harming quality, wasting the compute-in-memory advantage, and leaving hardware builders unsure whether a failure comes from the circuit or the degraded input. The paper introduces CQ-CiM, a unified hardware-aware framework. It jointly learns compression and quantization to produce low-bit vectors fit for diverse compute-in-memory designs. To the authors' knowledge, no earlier work shaped data for comprehensive compute-in-memory use in retrieval-augmented generation.
Catalog record
| CREATOR | Xinzhao Li* 1 , Alptekin Vardar* 2 , Franz Müller 2 , Navya Goli 3 , Umamaheswara Rao Tida 3 , Kai Ni 4 , X. Sharon Hu 4 , Thomas Kämpfe 2 , Ruiyang Qin 1 1 Villanova University, 2 Fraunhofer IPMS, 3 NDSU, 4 University of Notre Dame |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-20083