Provided is a memory system comprising an error correction code (ECC) decoder configured to receive data from a memory. The ECC decoder includes a syndrome generator configured to calculate at least one of syndrome vector and an erasure value, the calculation being devoid of erasure location information and an error-location polynomial generator configured to determine error location and error/erasure value polynomials responsive to syndrome and erasure calculation values output from the syndrome generator. An error value generator confirms error values at one or more known error locations based upon the determined error/erasure value polynomials, and an error location generator search for an error evaluation value to confirm the known error locations based upon the determined error location polynomials. Outputs of the error value generator and the error location generator are combined to produce corrected data.

Methods and apparatus are disclosed for managing the storage of static and dynamic neural network data within a non-volatile memory (NVM) die for use with deep neural networks (DNN). Some aspects relate to separate trim sets for separately configuring a static data NVM array for static input data and a dynamic data NVM array for dynamic synaptic weight data. For example, the static data NVM array may be configured via one trim set for data retention, whereas the dynamic data NVM array may be configured via another trim set for write performance. The trim sets may specify different configurations for error correction coding, write verification, and read threshold calibration, as well as different read/write voltage thresholds. In some examples, neural network regularization is provided within a DNN by setting trim parameters to encourage bit flips to avoid overfitting. Some examples relate to managing non-DNN data, such as stochastic gradient data.

A solution for detecting a multibyte error in a code word of a shortened error code is proposed, the shortened error code is a τ-byte-correcting error code, wherein bytes of the code word of the shortened error code determined a first range, the non-correctable multibyte error is detected if at least one of the following conditions is met: (a) at least one error position signal does not lie in the first range; (b) at least one error position signal indicates at least one error but fewer than terrors in the first range and no 1-byte error to (τ−1)-byte error is present.

A method for error correction comprises receiving data at a first device, and decoding, by decoder circuitry of the first device, the data. Decoding the data comprises determining a first error location within the data, and determining a first error magnitude within the data in parallel with determining the first error location. Decoding the data further comprises performing error correction to generate the decoded data based on the first error location and the first error magnitude. The method further comprises transmitting the decoded data to a second device.

A memory device can include a memory array, a processor coupled to the memory array, and a decoding apparatus. The decoding apparatus is configured to perform parallel decoding of codewords. Each of the codewords has a plurality of data blocks, each data block having a number of data bits. The decoding apparatus is configured to decode in parallel two or more codewords, which share a common data block, to determine error information associated with each codeword. For each error, the error information identifies a data block having the and associated error bit patterns. The decoding apparatus is configured to update the two or more codewords based on the identified data blocks having errors and the associated error bit patterns.

An error correction device according to the technical idea of the present disclosure includes a syndrome generation circuit configured to receive data and generate a plurality of syndromes for the data, a partial coefficient generation circuit configured to generate partial coefficient information on a part of a coefficient of an error location polynomial by using the data while the plurality of syndromes are generated, an error location determination circuit configured to determine the coefficient of the error location polynomial based on the plurality of syndromes and the partial coefficient information, and obtain a location of an error in the data by using the error location polynomial, and an error correction circuit configured to correct the error in the data according to the location of the error.

The present invention is directed to data communication systems and methods thereof. According to various embodiments, the present invention provides a communication with a reconfigurable forward-error-correction (FEC) module. The FEC module processes data received from two or more communication lanes, and depending on the mode of operation, the FEC module can combine data from the two or more communication lanes and perform error correction on the combined data, or the FEC module can processes data from the two communications lanes separately and perform error correction independently for the each of the data communication lanes. There are other embodiments as well.

A read method of a nonvolatile memory device is provided. The method includes storing data sensed from selected memory cells of the nonvolatile memory device into a page buffer, performing an error decoding operation by performing error detection on the sensed data to detect and error, correcting the detected error if the error is detected, and overwriting the page buffer with the corrected data, and de-randomizing data stored in the page buffer by using a seed after the error decoding operation has completed.

An ECC decoder includes a syndrome calculation block, a fast path controller, a KES block, a CSEE block, an UED, and a multiplexer. The KES block includes a plurality of KES-stages to calculate and output an error location/magnitude polynomial of a syndrome outputted from the syndrome calculation block. Each of a second to last KES-stages of the plurality of KES-stages receives the error location/magnitude polynomial from the previous KES-stage to output an error location/magnitude polynomial generated by an additional calculating operation. The additionally calculated error location/magnitude polynomial is not transmitted to the next KES-stage but directly outputted when an error location and an error magnitude are identified by the additionally calculated error location/magnitude polynomial.

An operating method of a controller includes generating error reliability of data based on reliability information of one or more error-corrected bits of the data, wherein the data is read out from a semiconductor memory device and a hard decision ECC decoding to the data through a BCH code is determined as successful; and determining miscorrection of the data based on the error reliability.