Hewlett Packard Enterprise Development LP

Catherine Graves

John Paul Strachan

Can Li

Giacomo Pedretti

Martin Foltin

Dejan S Milojicic

Suhas Kumar

Sergey Serebryakov

Paolo Faraboschi

Tobias Frederic Ziegler

Craig William Warner

Thomas Van Vaerenbergh

Sai Rahul Chalamalasetti

Amit S Sharma

Ron Roth

Pedro Henrique Rocha Bruel

Brent Buchanan

Kimberly Keeton

Le Zheng

Kivanc Ozonat

John Moon

Thomas Maurits M Van Vaerenbergh

Luca Buonanno

A technique for compressing an analog content addressable memory (CAM) array is provided. Random input data is applied to the analog CAM array, and an average measure of similarity is calculated for each output row of the analog CAM array. Rows of the analog CAM array that have measures of similarity that are close to each other can be eliminated, such as by removing similar rows or merging together similar rows. Thus, the analog CAM array size can be reduced without a loss in accuracy of a model stored on the analog CAM array.

Compression of analog content addressable memory

Systems and methods are provided for employing analog content addressable memory (aCAMs) to achieve low latency complex distribution sampling. For example, an aCAM core circuit can include an aCAM array. Amplitudes of a probability distribution function are mapped to a width of one or more aCAM cells in each row of the aCAM array. The aCAM core circuit can also include a resistive random access memory (RRAM) storing lookup information, such as information used for processing a model. By randomly selecting columns to search of the aCAM array, the mapped probability distribution function is sampled in a manner that has low latency. The aCAM core circuit can accelerate the sampling step in methods relying on sampling from arbitrary probability distributions, such as particle filter techniques. A hardware architecture for an aCAM Particle Filter that utilizes the aCAM core circuit as a central structure is also described.

CAMs for low latency complex distribution sampling

A system for facilitating an enhanced k-SAT solver is provided. The system can include a set of analog content addressable memory (aCAM) modules that can represent an expression in a conjunctive normal form (CNF), wherein a respective aCAM module corresponds to a clause of the expression. The system can also include a set of data lines that can provide input candidate values to the set of aCAM modules. A controller of the system can program the set of aCAM modules with respective analog values to represent the expression. The system can also include sensing logic block to determine a distance of a current solution from a target solution based on a combination of respective outputs from the set of aCAM modules. The controller can then iteratively modify an input value for a subset of data lines until the current solution converges based on a convergence condition.

Enhanced k-SAT solver using analog content addressable memory

The disclosure generally provides for a method of solving a K-SAT problem. The method comprises programming one or more clauses of a Boolean expression for a K-SAT problem written in negated disjunctive normal form (DNF) to a ternary-CAM (TCAM) array comprising columns and rows of TCAM cells, applying an interpretation comprising one or more binary variables expected to solve the Boolean expression as an input along the columns to the TCAM array, returning a binary value for each clause and updating one or more variables within the interpretation if at least one clause is violated.

Compact K-SAT verification with TCAMs

Systems and methods are provided for implementing a low power and area ternary content addressable memory (TCAM). An example of a TCAM comprises a match line, and a plurality of TCAM cells connected along the match line. Each TCAM cell stores a state of a threshold value. The TCAM cells are configured to pull down a signal over the match line in response to inequality between an input search and the threshold value. The plurality of TCAM cells comprises a number of TCAM cells that is less than the threshold value. The input values can be encoded according to a first encoding scheme and the threshold value can be encoded according to one of a second and a third encoding scheme based on an inequality check mapped to the plurality of TCAM cells.

Inequality check with ternary cam

Embodiments of the disclosure provide a system, method, or computer readable medium for providing a differentiable content addressable memory (aCAM) that implements an analog input analog storage and analog output learning memory. The analog output of the differentiable CAM can provide input to a learning algorithm, which may compute the gradients in comparison to historic values and reduce data inaccuracies and power consumption.

Differentiable content addressable memory

A crossbar array includes a number of memory elements. An analog-to-digital converter (ADC) is electronically coupled to the vector output register. A digital-to-analog converter (DAC) is electronically coupled to the vector input register. A processor is electronically coupled to the ADC and to the DAC. The processor may be configured to determine whether division of input vector data by output vector data from the crossbar array is within a threshold value, and if not within the threshold value, determine changed data values as between the output vector data and the input vector data, and write the changed data values to the memory elements of the crossbar array.

Acceleration of model/weight programming in memristor crossbar arrays

Embodiments of the disclosure provide a system, method, or computer readable medium for programming a target analog voltage range of an analog content addressable memory (aCAM) row. The method may comprise calculating a threshold current sufficient to switch a sense amplifier (SA) on and discharge a match line (ML) connected to a cell of the aCAM; and based on calculating the threshold current, programming a match threshold value by setting a memristor conductance in association with the target analog voltage range applied to a data line (DL) input. The target analog voltage range may comprise a target analog voltage range vector.

Iterative programming of analog content addressable memory

The disclosure generally provides for a method of solving a K-SAT problem. The method comprises programming one or more clauses of a Boolean expression for a K-SAT problem written in negated disjunctive normal form (DNF) to a ternary-CAM (TCAM) array comprising columns and rows of TCAM cells. The method further includes applying an interpretation comprising one or more binary variables expected to solve the Boolean expression as an input along the columns to the TCAM array, returning a binary value for each clause, randomly selecting one matched match line, determining a selected clause from one or more violated clause, and altering one or more literals within the interpretation using a break count for each variable of the selected clause.

Stochastic local search accelerator based on cams for solving constraint satisfaction problems

Examples increase precision for aCAMs by converting an input signal (x) received by a circuit into a first analog voltage signal (V(x)) representing the most significant bits of the input signal (x) and a second analog voltage signal (V(x)) representing the least significant bits of the input signal (x). By dividing the input signal (x) bit-wise into the first analog voltage signal (V(x)) and the second analog voltage signal (V(x)), the circuit can utilize aCAM sub-circuits implementing a combination of Boolean operations to search the input signal (x) against 2programmable levels, where 'M' represents the number of programmable bits for each aCAM sub-circuit. Thus, using similar circuit hardware, example circuits square the number of programmable levels of conventional aCAMs (which generally only have 2programmable levels). Accordingly, examples provide new aCAMs that can carry out more complex computations than conventional aCAMs of comparable cost, size, and power consumption.

Increased precision analog content addressable memories

Systems are methods are provided for implementing an analog content addressable memory (analog CAM), which is particularly structured to allow for an amount of variance (fuzziness) in its search operations. The analog CAM may search for approximate matches with the data stored therein, or matches within a defined variance. Circuitry of the analog CAM may include transistor-source lines that receive search-variance parameters, and/or data lines that receive search-variance parameters explicitly within the search input data. The search-variance parameters may include an upper bound and a lower bound that define a range of values within the allotted amount of fuzziness (e.g., deviation from the stored value). The search-variance parameters may program (using analog approaches) the analog CAM to perform searches having a modifiable restrictiveness that is tuned dynamically, as defined by the input search-variance. Thus, highly efficient hardware for complex applications involving fuzziness are enabled.

Methods and systems for an analog cam with fuzzy search

An analog error correction circuit is disclosed that implements an analog error correction code. The analog circuit includes a crossbar array of memristors or other non-volatile tunable resistive memory devices. The crossbar array includes a first crossbar array portion programmed with values of a target computation matrix and a second crossbar array portion programmed with values of an encoder matrix for correcting computation errors in the matrix multiplication of an input vector with the computation matrix. The first and second crossbar array portions share the same row lines and are connected to a third crossbar array portion that is programmed with values of a decoder matrix, thereby enabling single-cycle error detection. A computation error is detected based on output of the decoder matrix circuitry and a location of the error is determined via an inverse matrix multiplication operation whereby the decoder matrix output is fed back to the decoder matrix.

Analog error detection and correction in analog in-memory crossbars

Examples of the presently disclosed technology provide new circuits for detecting errors in aCAMs with improved efficiency. Specifically designed around the structure and operation of aCAM arrays, these circuits include counter sub-circuits electrically connected to match lines of aCAM rows such that the counter sub-circuits receive match-related signals output from aCAM rows. The value stored by a counter sub-circuit may change in response to receiving a match signal, and may remain the same in response to receiving a mismatch signal. As will be described in greater detail below, the stored value of the counter sub-circuit may be used to detect/identify an error in its associated aCAM row after a set of (specially-computed) error-detection input vectors are sequentially applied to the circuit.

Growing community of inventors

Palo Alto, CA, United States of America

Catherine Graves