Table 2 Representative works on error mitigation in AIMC systems and quantitative benchmarks
From: Achieving high precision in analog in-memory computing systems
Authors | Year | Technique | Main result |
|---|---|---|---|
Hu et al.77 | 2016 | Slicing | 7-bit computation on 4-bit primitive by binary slicing (+3 bits) |
Shafiee et al.76 | 2016 | Slicing | 16-bit computation by binary matrix and input slicing (+15 bits) |
Ankit et al.80 | 2019 | Slicing | 16-bit computation by binary matrix and input slicing (+15 bits) |
Pedretti et al.79 | 2021 | Slicing | 11-bit weight precision on 4-bit memory by analog slicing (+7 bits) |
Le Gallo et al.75 | 2022 | Slicing | Binary slicing optimization for bit-width and drift mitigation |
Khaddam-Aljameh et al.13 | 2022 | Slicing | 7-bit input slicing by PWM with time-to-digital readout |
Jiang et al.74 | 2022 | Slicing | Analog input slicing with capacitive readout and PWM generation |
Song et al.21 | 2024 | Slicing | Arbitrary precision on 3-bit memory by analog slicing (≈ +13 bits) |
Salamat et al.94 | 2018 | RNS | Simulated 16-bit accuracy on 6-bit analog CBA (+10 bits) |
Roohi et al.95 | 2021 | RNS | Simulated 16-bit accuracy on 5-bit digital CBA (+11 bits) |
Demirkiran et al.85 | 2024 | RNS | FP32 inference accuracy on 6-bit analog primitives (≈ +18 bit) |
Niu et al.107 | 2012 | ECC | Bit-level CBA protection by Hamming code-based ECC |
Feinberg et al.109 | 2018 | ECC | 16-bit compute-level protection by ABN ECC |
Li, Roth et al.66 | 2020 | ECC | Correction of outlying errors by fully-analog ECC |
Li, Read et al.110 | 2022 | ECC | In situ analog/digital AN code-based ECC for MSB correction |
Parrini et al.136 | 2024 | ECC | 4-bit weight protection by analog ECC |
Le Gallo et al.49 | 2018 | IRF | 10−6 solution accuracy with 4-bit AIMC primitive (+16 bits) |
Feinberg et al.125 | 2021 | IRF | In situ preconditioning by AIMC IMVM |
Kalantzis, Gupta et al.127 | 2021 | IRF | AIMC-preconditioned linear system solution by Richardson iteration |
Wu et al.129 | 2023 | IRF | AIMC-tailored iterative update algorithm |
Kalantzis, Squillante et al.128 | 2023 | IRF | AIMC-preconditioned linear system solution by GMRES algorithm |
Li, Xue et al.124 | 2025 | IRF | 10−10 accuracy with fully-analog iterative AIMC solver |
