arXiv subjects

Xiaobo Sharon Hu

Publications and source records attributed to Xiaobo Sharon Hu.

At least 19 recordsLinked to original sources

arXiv2026

Interface-Aware KV Cache Quantization for Dense On-Chip NVM in Long-Context LLM Decoding

The key-value (KV) cache is the dominant memory bottleneck in long-context large language model (LLM) decoding: every step reads it entirely, so decoding is memory-bandwidth bound. Holding a quantized KV cache in dense on-chip non-volatile memory (NVM) removes the off-chip transfer. Existing KV quantization methods, however, were designed for GPU-style memory systems: KIVI attaches per-group metadata, adding about 25% to the stored KV cache; KVQuant keeps sparse full-precision outliers that a dense array cannot hold in place. This paper examines what these structures cost when the KV cache resides in NVM behind fixed-range converters, and designs a quantization scheme matched to that interface. The architecture stores the quantized KV cache in dense on-chip NVM, uses a small static analog crossbar only for the fixed rotation, and keeps attention in on-chip digital logic. A randomized rotation and per-vector normalization give every coordinate the same range, so one fixed codebook for keys and one for values, each shared across all tokens of the corresponding tensor type, serve the entire KV cache. The codebook thresholds are programmed once as the read converter's reference levels, enabling fixed-range digitization with no per-token converter reconfiguration. Dequantization is a sixteen-entry lookup and one norm multiply; the only per-vector metadata is one scalar, about 3%. Across models from 3B to 14B and contexts to 32k tokens, the four-bit KV cache maintains accuracy under storage and crossbar noise simulated at realistic device levels. KIVI and KVQuant remain more accurate in software; the advantage of our format lies at the memory interface: 3.1-3.6x lower KV read energy than both mapped to the same NVM, and 8x lower metadata overhead than KIVI. The contribution is a KV quantization co-designed with the NVM memory interface rather than a new accuracy record.

cs.AR↗

arXiv2026

PolySim: Deterministic Polynomial Surrogates for Cross-Modal Retrieval on CiM

Cross-modal retrieval on edge devices benefits from probabilistic embeddings that capture semantic uncertainty, but deploying them on compute-in-memory (CiM) hardware remains an open problem. The core difficulty is a sampling gap: probabilistic methods such as PCME rely on Monte Carlo sampling and nonlinear distance evaluation at inference, which are fundamentally incompatible with CiM crossbar arrays that support only deterministic, single-step matrix-vector multiplication. Few existing probabilistic retrieval methods can be executed on a conventional crossbar. To bridge this gap, we propose PolySim, a framework that reformulates probabilistic retrieval into a fully deterministic pipeline. PolySim approximates each Gaussian embedding dimension using low-order polynomial bases and computes similarity via a learnable order-bilinear kernel, eliminating stochastic sampling while preserving distributional information. In experiments on six benchmarks spanning video, image, and audio retrieval, PolySim improves R@1 over deterministic baselines by up to 10.3\% and matches or exceeds PCME, while reducing inference to a single crossbar-compatible matrix-vector multiplication. CrossSim evaluation under realistic device non-idealities confirms robust deployment on conventional crossbar arrays. To the best of our knowledge, PolySim is the first method to enable probabilistic cross-modal retrieval on CiM hardware.

cs.ET↗

arXiv2026

CQ-CiM: Hardware-Aware Embedding Shaping for Robust CiM-Based Retrieval

Deploying Retrieval-Augmented Generation (RAG) on edge devices is in high demand, but is hindered by the latency of massive data movement and computation on traditional architectures. Compute-in-Memory (CiM) architectures address this bottleneck by performing vector search directly within their crossbar structure. However, CiM's adoption for RAG is limited by a fundamental ``representation gap,'' as high-precision, high-dimension embeddings are incompatible with CiM's low-precision, low-dimension array constraints. This gap is compounded by the diversity of CiM implementations (e.g., SRAM, ReRAM, FeFET), each with unique designs (e.g., 2-bit cells, 512x512 arrays). Consequently, RAG data must be naively reshaped to fit each target implementation. Current data shaping methods handle dimension and precision disjointly, which degrades data fidelity. This not only negates the advantages of CiM for RAG but also confuses hardware designers, making it unclear if a failure is due to the circuit design or the degraded input data. As a result, CiM adoption remains limited. In this paper, we introduce CQ-CiM, a unified, hardware-aware data shaping framework that jointly learns Compression and Quantization to produce CiM-compatible low-bit embeddings for diverse CiM designs. To the best of our knowledge, this is the first work to shape data for comprehensive CiM usage on RAG.

cs.ET↗

arXiv2026

When Small Variations Become Big Failures: Reliability Challenges in Compute-in-Memory Neural Accelerators

Compute-in-memory (CiM) architectures promise significant improvements in energy efficiency and throughput for deep neural network acceleration by alleviating the von Neumann bottleneck. However, their reliance on emerging non-volatile memory devices introduces device-level non-idealities-such as write variability, conductance drift, and stochastic noise-that fundamentally challenge reliability, predictability, and safety, especially in safety-critical applications. This talk examines the reliability limits of CiM-based neural accelerators and presents a series of techniques that bridge device physics, architecture, and learning algorithms to address these challenges. We first demonstrate that even small device variations can lead to disproportionately large accuracy degradation and catastrophic failures in safety-critical inference workloads, revealing a critical gap between average-case evaluations and worst-case behavior. Building on this insight, we introduce SWIM, a selective write-verify mechanism that strategically applies verification only where it is most impactful, significantly improving reliability while maintaining CiM's efficiency advantages. Finally, we explore a learning-centric solution that improves realistic worst-case performance by training neural networks with right-censored Gaussian noise, aligning training assumptions with hardware-induced variability and enabling robust deployment without excessive hardware overhead. Together, these works highlight the necessity of cross-layer co-design for CiM accelerators and provide a principled path toward dependable, efficient neural inference on emerging memory technologies-paving the way for their adoption in safety- and reliability-critical systems.

cs.LG↗

arXiv2026

Secure Scattered Memory: Rethinking Secure Enclave Memory with Secret Sharing

The rise of cloud computing demands secure memory systems that ensure data confidentiality, integrity, and freshness against replay attacks. Existing schemes such as AES-XTS, AES-GCM, and AES-CTR each trade performance for security, with only AES-CTR plus Message Authentication Codes (MAC) and Merkle Trees (MT) providing full protection - at the cost of substantial counter and MT overhead. This paper introduces Secure Scattered Memory (SSM), a novel scheme that replaces counter-based encryption with polynomial-based secret sharing. Each data block is encoded into multiple cryptographically independent shares distributed across memory, inherently preventing information leakage while ensuring integrity and freshness through mathematical reconstruction properties. Implemented and synthesized in a 28 nm commercial PDK, SSM occupies 0.27 mm^2 and consumes 284.53 mW. Experiments show only 10% and 8% performance overhead over AES-XTS and AES-GCM, respectively, while outperforming Morphable Counter (MICRO 2018) by up to 40%, achieving 12% better performance than EMCC/RMCC (MICRO 2022), and exceeding COSMOS (MICRO 2025) by 3%.

cs.CR↗

arXiv2025

Special Session: Sustainable Deployment of Deep Neural Networks on Non-Volatile Compute-in-Memory Accelerators

Non-volatile memory (NVM) based compute-in-memory (CIM) accelerators have emerged as a sustainable solution to significantly boost energy efficiency and minimize latency for Deep Neural Networks (DNNs) inference due to their in-situ data processing capabilities. However, the performance of NVCIM accelerators degrades because of the stochastic nature and intrinsic variations of NVM devices. Conventional write-verify operations, which enhance inference accuracy through iterative writing and verification during deployment, are costly in terms of energy and time. Inspired by negative feedback theory, we present a novel negative optimization training mechanism to achieve robust DNN deployment for NVCIM. We develop an Oriented Variational Forward (OVF) training method to implement this mechanism. Experiments show that OVF outperforms existing state-of-the-art techniques with up to a 46.71% improvement in inference accuracy while reducing epistemic uncertainty. This mechanism reduces the reliance on write-verify operations and thus contributes to the sustainable and practical deployment of NVCIM accelerators, addressing performance degradation while maintaining the benefits of sustainable computing with NVCIM accelerators.

cs.AR↗

arXiv2025

NeFT: Negative Feedback Training to Improve Robustness of Compute-In-Memory DNN Accelerators

Compute-in-memory accelerators built upon non-volatile memory devices excel in energy efficiency and latency when performing deep neural network (DNN) inference, thanks to their in-situ data processing capability. However, the stochastic nature and intrinsic variations of non-volatile memory devices often result in performance degradation during DNN inference. Introducing these non-ideal device behaviors in DNN training enhances robustness, but drawbacks include limited accuracy improvement, reduced prediction confidence, and convergence issues. This arises from a mismatch between the deterministic training and non-deterministic device variations, as such training, though considering variations, relies solely on the model's final output. In this work, inspired by control theory, we propose Negative Feedback Training (NeFT), a novel concept supported by theoretical analysis, to more effectively capture the multi-scale noisy information throughout the network. We instantiate this concept with two specific instances, oriented variational forward (OVF) and intermediate representation snapshot (IRS). Based on device variation models extracted from measured data, extensive experiments show that our NeFT outperforms existing state-of-the-art methods with up to a 45.08% improvement in inference accuracy while reducing epistemic uncertainty, boosting output confidence, and improving convergence probability. These results underline the generality and practicality of our NeFT framework for increasing the robustness of DNNs against device variations. The source code for these two instances is available at https://github.com/YifanQin-ND/NeFT_CIM

cs.LG↗

arXiv2025

Shared-PIM: Enabling Concurrent Computation and Data Flow for Faster Processing-in-DRAM

Processing-in-Memory (PIM) enhances memory with computational capabilities, potentially solving energy and latency issues associated with data transfer between memory and processors. However, managing concurrent computation and data flow within the PIM architecture incurs significant latency and energy penalty for applications. This paper introduces Shared-PIM, an architecture for in-DRAM PIM that strategically allocates rows in memory banks, bolstered by memory peripherals, for concurrent processing and data movement. Shared-PIM enables simultaneous computation and data transfer within a memory bank. When compared to LISA, a state-of-the-art architecture that facilitates data transfers for in-DRAM PIM, Shared-PIM reduces data movement latency and energy by 5x and 1.2x respectively. Furthermore, when integrated to a state-of-the-art (SOTA) in-DRAM PIM architecture (pLUTo), Shared-PIM achieves 1.4x faster addition and multiplication, and thereby improves the performance of matrix multiplication (MM) tasks by 40%, polynomial multiplication (PMM) by 44%, and numeric number transfer (NTT) tasks by 31%. Moreover, for graph processing tasks like Breadth-First Search (BFS) and Depth-First Search (DFS), Shared-PIM achieves a 29% improvement in speed, all with an area overhead of just 7.16% compared to the baseline pLUTo.

cs.AR↗

arXiv2025

TAP-CAM: A Tunable Approximate Matching Engine based on Ferroelectric Content Addressable Memory

Pattern search is crucial in numerous analytic applications for retrieving data entries akin to the query. Content Addressable Memories (CAMs), an in-memory computing fabric, directly compare input queries with stored entries through embedded comparison logic, facilitating fast parallel pattern search in memory. While conventional CAM designs offer exact match functionality, they are inadequate for meeting the approximate search needs of emerging data-intensive applications. Some recent CAM designs propose approximate matching functions, but they face limitations such as excessively large cell area or the inability to precisely control the degree of approximation. In this paper, we propose TAP-CAM, a novel ferroelectric field effect transistor (FeFET) based ternary CAM (TCAM) capable of both exact and tunable approximate matching. TAP-CAM employs a compact 2FeFET-2R cell structure as the entry storage unit, and similarities in Hamming distances between input queries and stored entries are measured using an evaluation transistor associated with the matchline of CAM array. The operation, robustness and performance of the proposed design at array level have been discussed and evaluated, respectively. We conduct a case study of K-nearest neighbor (KNN) search to benchmark the proposed TAP-CAM at application level. Results demonstrate that compared to 16T CMOS CAM with exact match functionality, TAP-CAM achieves a 16.95x energy improvement, along with a 3.06% accuracy enhancement. Compared to 2FeFET TCAM with approximate match functionality, TAP-CAM achieves a 6.78x energy improvement.

cs.ET↗

arXiv2024

A Remedy to Compute-in-Memory with Dynamic Random Access Memory: 1FeFET-1C Technology for Neuro-Symbolic AI

Neuro-symbolic artificial intelligence (AI) excels at learning from noisy and generalized patterns, conducting logical inferences, and providing interpretable reasoning. Comprising a 'neuro' component for feature extraction and a 'symbolic' component for decision-making, neuro-symbolic AI has yet to fully benefit from efficient hardware accelerators. Additionally, current hardware struggles to accommodate applications requiring dynamic resource allocation between these two components. To address these challenges-and mitigate the typical data-transfer bottleneck of classical Von Neumann architectures-we propose a ferroelectric charge-domain compute-in-memory (CiM) array as the foundational processing element for neuro-symbolic AI. This array seamlessly handles both the critical multiply-accumulate (MAC) operations of the 'neuro' workload and the parallel associative search operations of the 'symbolic' workload. To enable this approach, we introduce an innovative 1FeFET-1C cell, combining a ferroelectric field-effect transistor (FeFET) with a capacitor. This design, overcomes the destructive sensing limitations of DRAM in CiM applications, while capable of capitalizing decades of DRAM expertise with a similar cell structure as DRAM, achieves high immunity against FeFET variation-crucial for neuro-symbolic AI-and demonstrates superior energy efficiency. The functionalities of our design have been successfully validated through SPICE simulations and prototype fabrication and testing. Our hardware platform has been benchmarked in executing typical neuro-symbolic AI reasoning tasks, showing over 2x improvement in latency and 1000x improvement in energy efficiency compared to GPU-based implementations.

cs.ET↗

arXiv2024

TSB: Tiny Shared Block for Efficient DNN Deployment on NVCIM Accelerators

Compute-in-memory (CIM) accelerators using non-volatile memory (NVM) devices offer promising solutions for energy-efficient and low-latency Deep Neural Network (DNN) inference execution. However, practical deployment is often hindered by the challenge of dealing with the massive amount of model weight parameters impacted by the inherent device variations within non-volatile computing-in-memory (NVCIM) accelerators. This issue significantly offsets their advantages by increasing training overhead, the time and energy needed for mapping weights to device states, and diminishing inference accuracy. To mitigate these challenges, we propose the "Tiny Shared Block (TSB)" method, which integrates a small shared 1x1 convolution block into the DNN architecture. This block is designed to stabilize feature processing across the network, effectively reducing the impact of device variation. Extensive experimental results show that TSB achieves over 20x inference accuracy gap improvement, over 5x training speedup, and weights-to-device mapping cost reduction while requiring less than 0.4% of the original weights to be write-verified during programming, when compared with state-of-the-art baseline solutions. Our approach provides a practical and efficient solution for deploying robust DNN models on NVCIM accelerators, making it a valuable contribution to the field of energy-efficient AI hardware.

cs.AR↗

arXiv2024

A Computing-in-Memory-based One-Class Hyperdimensional Computing Model for Outlier Detection

In this work, we present ODHD, an algorithm for outlier detection based on hyperdimensional computing (HDC), a non-classical learning paradigm. Along with the HDC-based algorithm, we propose IM-ODHD, a computing-in-memory (CiM) implementation based on hardware/software (HW/SW) codesign for improved latency and energy efficiency. The training and testing phases of ODHD may be performed with conventional CPU/GPU hardware or our IM-ODHD, SRAM-based CiM architecture using the proposed HW/SW codesign techniques. We evaluate the performance of ODHD on six datasets from different application domains using three metrics, namely accuracy, F1 score, and ROC-AUC, and compare it with multiple baseline methods such as OCSVM, isolation forest, and autoencoder. The experimental results indicate that ODHD outperforms all the baseline methods in terms of these three metrics on every dataset for both CPU/GPU and CiM implementations. Furthermore, we perform an extensive design space exploration to demonstrate the tradeoff between delay, energy efficiency, and performance of ODHD. We demonstrate that the HW/SW codesign implementation of the outlier detection on IM-ODHD is able to outperform the GPU-based implementation of ODHD by at least 331.5x/889x in terms of training/testing latency (and on average 14.0x/36.9x in terms of training/testing energy consumption.

cs.AR↗

arXiv2023

Compute-in-Memory based Neural Network Accelerators for Safety-Critical Systems: Worst-Case Scenarios and Protections

Emerging non-volatile memory (NVM)-based Computing-in-Memory (CiM) architectures show substantial promise in accelerating deep neural networks (DNNs) due to their exceptional energy efficiency. However, NVM devices are prone to device variations. Consequently, the actual DNN weights mapped to NVM devices can differ considerably from their targeted values, inducing significant performance degradation. Many existing solutions aim to optimize average performance amidst device variations, which is a suitable strategy for general-purpose conditions. However, the worst-case performance that is crucial for safety-critical applications is largely overlooked in current research. In this study, we define the problem of pinpointing the worst-case performance of CiM DNN accelerators affected by device variations. Additionally, we introduce a strategy to identify a specific pattern of the device value deviations in the complex, high-dimensional value deviation space, responsible for this worst-case outcome. Our findings reveal that even subtle device variations can precipitate a dramatic decline in DNN accuracy, posing risks for CiM-based platforms in supporting safety-critical applications. Notably, we observe that prevailing techniques to bolster average DNN performance in CiM accelerators fall short in enhancing worst-case scenarios. In light of this issue, we propose a novel worst-case-aware training technique named A-TRICE that efficiently combines adversarial training and noise-injection training with right-censored Gaussian noise to improve the DNN accuracy in the worst-case scenarios. Our experimental results demonstrate that A-TRICE improves the worst-case accuracy under device variations by up to 33%.

cs.LG↗

arXiv2023

U-SWIM: Universal Selective Write-Verify for Computing-in-Memory Neural Accelerators

Architectures that incorporate Computing-in-Memory (CiM) using emerging non-volatile memory (NVM) devices have become strong contenders for deep neural network (DNN) acceleration due to their impressive energy efficiency. Yet, a significant challenge arises when using these emerging devices: they can show substantial variations during the weight-mapping process. This can severely impact DNN accuracy if not mitigated. A widely accepted remedy for imperfect weight mapping is the iterative write-verify approach, which involves verifying conductance values and adjusting devices if needed. In all existing publications, this procedure is applied to every individual device, resulting in a significant programming time overhead. In our research, we illustrate that only a small fraction of weights need this write-verify treatment for the corresponding devices and the DNN accuracy can be preserved, yielding a notable programming acceleration. Building on this, we introduce USWIM, a novel method based on the second derivative. It leverages a single iteration of forward and backpropagation to pinpoint the weights demanding write-verify. Through extensive tests on diverse DNN designs and datasets, USWIM manifests up to a 10x programming acceleration against the traditional exhaustive write-verify method, all while maintaining a similar accuracy level. Furthermore, compared to our earlier SWIM technique, USWIM excels, showing a 7x speedup when dealing with devices exhibiting non-uniform variations.

cs.AR↗

arXiv2023

C4CAM: A Compiler for CAM-based In-memory Accelerators

Machine learning and data analytics applications increasingly suffer from the high latency and energy consumption of conventional von Neumann architectures. Recently, several in-memory and near-memory systems have been proposed to remove this von Neumann bottleneck. Platforms based on content-addressable memories (CAMs) are particularly interesting due to their efficient support for the search-based operations that form the foundation for many applications, including K-nearest neighbors (KNN), high-dimensional computing (HDC), recommender systems, and one-shot learning among others. Today, these platforms are designed by hand and can only be programmed with low-level code, accessible only to hardware experts. In this paper, we introduce C4CAM, the first compiler framework to quickly explore CAM configurations and to seamlessly generate code from high-level TorchScript code. C4CAM employs a hierarchy of abstractions that progressively lowers programs, allowing code transformations at the most suitable abstraction level. Depending on the type and technology, CAM arrays exhibit varying latencies and power profiles. Our framework allows analyzing the impact of such differences in terms of system-level performance and energy consumption, and thus supports designers in selecting appropriate designs for a given application.

cs.AR↗

arXiv2023

Privacy Preserving In-memory Computing Engine

Privacy has rapidly become a major concern/design consideration. Homomorphic Encryption (HE) and Garbled Circuits (GC) are privacy-preserving techniques that support computations on encrypted data. HE and GC can complement each other, as HE is more efficient for linear operations, while GC is more effective for non-linear operations. Together, they enable complex computing tasks, such as machine learning, to be performed exactly on ciphertexts. However, HE and GC introduce two major bottlenecks: an elevated computational overhead and high data transfer costs. This paper presents PPIMCE, an in-memory computing (IMC) fabric designed to mitigate both computational overhead and data transfer issues. Through the use of multiple IMC cores for high parallelism, and by leveraging in-SRAM IMC for data management, PPIMCE offers a compact, energy-efficient solution for accelerating HE and GC. PPIMCE achieves a 107X speedup against a CPU implementation of GC. Additionally, PPIMCE achieves a 1,500X and 800X speedup compared to CPU and GPU implementations of CKKS-based HE multiplications. For privacy-preserving machine learning inference, PPIMCE attains a 1,000X speedup compared to CPU and a 12X speedup against CraterLake, the state-of-art privacy preserving computation accelerator.

cs.CR↗

arXiv2023

Improving Realistic Worst-Case Performance of NVCiM DNN Accelerators through Training with Right-Censored Gaussian Noise

Compute-in-Memory (CiM), built upon non-volatile memory (NVM) devices, is promising for accelerating deep neural networks (DNNs) owing to its in-situ data processing capability and superior energy efficiency. Unfortunately, the well-trained model parameters, after being mapped to NVM devices, can often exhibit large deviations from their intended values due to device variations, resulting in notable performance degradation in these CiM-based DNN accelerators. There exists a long list of solutions to address this issue. However, they mainly focus on improving the mean performance of CiM DNN accelerators. How to guarantee the worst-case performance under the impact of device variations, which is crucial for many safety-critical applications such as self-driving cars, has been far less explored. In this work, we propose to use the k-th percentile performance (KPP) to capture the realistic worst-case performance of DNN models executing on CiM accelerators. Through a formal analysis of the properties of KPP and the noise injection-based DNN training, we demonstrate that injecting a novel right-censored Gaussian noise, as opposed to the conventional Gaussian noise, significantly improves the KPP of DNNs. We further propose an automated method to determine the optimal hyperparameters for injecting this right-censored Gaussian noise during the training process. Our method achieves up to a 26% improvement in KPP compared to the state-of-the-art methods employed to enhance DNN robustness under the impact of device variations.

cs.LG↗

arXiv2023

Accelerating Polynomial Modular Multiplication with Crossbar-Based Compute-in-Memory

Lattice-based cryptographic algorithms built on ring learning with error theory are gaining importance due to their potential for providing post-quantum security. However, these algorithms involve complex polynomial operations, such as polynomial modular multiplication (PMM), which is the most time-consuming part of these algorithms. Accelerating PMM is crucial to make lattice-based cryptographic algorithms widely adopted by more applications. This work introduces a novel high-throughput and compact PMM accelerator, X-Poly, based on the crossbar (XB)-type compute-in-memory (CIM). We identify the most appropriate PMM algorithm for XB-CIM. We then propose a novel bit-mapping technique to reduce the area and energy of the XB-CIM fabric, and conduct processing engine (PE)-level optimization to increase memory utilization and support different problem sizes with a fixed number of XB arrays. X-Poly design achieves 3.1X10^6 PMM operations/s throughput and offers 200X latency improvement compared to the CPU-based implementation. It also achieves 3.9X throughput per area improvement compared with the state-of-the-art CIM accelerators.

cs.CR↗