Apple Inc.

Ramot at Tel-aviv University Ltd.

Runai Labs Ltd.

Ronen Dar

Micha Anholt

Tomer Ish-Shalom

Naftali Sommer

Ofir Shalvi

Eyal Gurgi

Meir Feder

Avraham Poza Meir

Uri Perlmutter

Barak Baum

Mark Shtaif

Yael Shur

Antonio Mecozzi

Tal Inbar

Alex Radinski

A method for training a Neural-Network (NN), the method includes receiving a plurality of NN training tasks, each training task including (i) a respective preprocessing phase that preprocesses data to be provided as input data to the NN, and (ii) a respective computation phase that trains the NN using the preprocessed data. The plurality of NN training tasks is executed, including: (a) a commonality is identified between the input data required by computation phases of two or more of the training tasks, and (b) in response to identifying the commonality, one or more preprocessing phases are executed that produce the input data jointly for the two or more training tasks.

Sharing preprocessing, computations, and hardware resources between multiple neural networks

A method that may include receiving a block of signals from a certain wavelength division multiplex (WDM) channel out of a set of WDM channels; analyzing at least a first sub-block of signals of the block of signals to provide analysis results indicative of interferences that affect the first sub-block of signals and result from transmissions over other WDM channels of the set of WDM channels; and mitigating interferences that affect the block of signals in response to the analysis results.

Nonlinear noise cancellation in wavelength division multiplexing (WDM) systems

A method includes storing data in a memory that includes multiple analog memory cells. After storing the data, an interference caused by a first group of the analog memory cells to a second group of the analog memory cells is estimated. The data stored in the first group is reconstructed based on the estimated interference caused by the first group to the second group.

Readout of interfering memory cells using estimated interference to other memory cells

A method includes decoding a code word of an Error Correction Code (ECC), which is representable by a set of check equations, by performing a sequence of iterations, such that each iteration involves processing of multiple variable nodes. For one or more selected variable nodes, a count of the check equations that are defined over one or more variables held respectively by the one or more selected variable nodes is evaluated, and, when the count meets a predefined skipping criterion, the one or more selected variable nodes are omitted from a given iteration in the sequence.

Decoder with selective iteration scheduling

A method includes programming a group of analog memory cells by writing respective analog values into the memory cells in the group. After programming the group, the analog values are read from the memory cells in the group using a set of read thresholds so as to produce readout results. Respective optimal positions for the read thresholds in the set are identified based on the readout results. A noise level in the readout results is estimated based on the identified optimal positions of the read thresholds.

SNR estimation in analog memory cells

A method includes storing data in a group of analog memory cells by writing respective analog values into the memory cells in the group. After storing the data, the analog values are read from the memory cells in the group one or more times using one or more respective read thresholds so as to produce readout results. Reliability measures are computed for the read analog values based on the readout results. An offset of the one or more read thresholds from an optimal read threshold position is estimated based on the reliability measures. The reliability measures are modified to compensate for the estimated offset, and the data stored in the analog memory cells in the group is decoded using the corrected reliability measures.

Soft readout from analog memory cells in the presence of read threshold errors

Signal-to-Noise Ratio (SNR) estimation in analog memory cells based on optimal read thresholds

A method and system for transmitting data over a Jacobi MIMO channel when using channel state feedback.

System and method for exchanging information over a jacobi MIMO channel

A method includes, in an Error Correction Code (ECC) decoder that includes variable nodes and check nodes, receiving in a given variable node Check-to-Variable (C2V) messages from a subset of the check nodes. Magnitudes for respective Variable-to-Check (V2C) messages to be sent to the check nodes in the subset are computed based on the received C2V messages. A single sign is computed, for use in all the V2C messages to be sent from the given variable node to the check nodes in the subset. The V2C messages are sent from the given variable node to the check nodes in the subset, such that each V2C message includes a respective magnitude and the single sign.

Sign equalization in iterative decoding of sparse graph codes

A method for data storage includes accepting data for storage in a memory including multiple analog memory cells. For each memory cell, a respective set of nominal analog values is assigned for representing data values to be stored in the memory cell, by choosing the nominal analog values for a given memory cell in a respective range that depends on interference between the given memory cell and at least one other memory cell in the memory. The data is stored in each memory cell using the respective selected set of the nominal analog values.

Interference-aware assignment of programming levels in analog memory cells

A method includes performing a read operation that reads data from a group of analog memory cells using at least one read threshold, to produce readout results. A detection is made that the read threshold is set in a restricted range that causes the readout results not to reflect the read threshold. The data is reproduced from the group of the memory cells while compensating for the read threshold that is set in the restricted range.

Identifying and mitigating restricted sampling voltage ranges in analog memory cells

A method for data storage includes storing data in analog memory cells by programming the memory cells with respective analog input values. After storing the data, respective analog output values are read from the memory cells using multiple read thresholds, which define multiple ranges of the analog output values. Respective numbers of read errors in the data, corresponding to the analog output values falling in the ranges, are assessed. The stored data is recovered based on respective numbers of the read errors assessed in the ranges.

Read threshold setting based on soft readout statistics

A method for data storage includes accepting data for storage in a memory that includes multiple analog memory cells. The data is stored in a first group of the memory cells by programming a second group of the memory cells so as to cause the second group to generate interference in the first group, and individually erasing the first group while verifying that analog levels of the memory cells in the first group subject to the interference are within a predefined bound following erasure. After erasing the first group, the first group of the memory cells is programmed with the data.

Growing community of inventors

Tel Aviv, Israel

Ronen Dar