TY  - JOUR
T1  - Selective AN Code Design for Correcting Computation Errors and Mis-Decoding Detection in Process-in-Memory
AU  - Kim, Myung-in 
AU  - Ha, Jeong-seok 
JO  - The Journal of Korean Institute of Communications and Information Sciences
PY  - 2024
DA  - 2024/1/1
DO  - 10.7840/kics.2024.49.1.104
KW  - PIM
KW  - AN codes
KW  - weight matrix
KW  - look-up-table
KW  - correctable set
KW  - mis-decoding
AB  - This work proposes a selective AN coding scheme for correcting errors which happen during computations in the process-in-memory devices. AN codes are error-correcting codes which encode the message data by multiplying a positive integer A to them. AN codes correct errors by mapping the division remainders of the erroneous coded data divided by A to the error patterns in the look-up-table. This paper proposes an AN coding scheme which selectively determines different error-correcting capabilities according to the significancy of computational results. At first, the proposed work defines the set of bit-lines with high significance as the correctable set. Then, it identifies conditions which ensure the correction and the mis-decoding detection for the error patterns inside and outside of the correctable set, respectively. The proposed AN coding scheme enables more flexible code design for a given noisy environment compared to the existing static AN coding scheme.
Additionally, in terms of the storage capacity, a more efficient decoder implementation is possible compared to the ABN coding scheme which is known suitable for flexible code design.