Robert Boschgmbh

Andre Guntoro

Heiner Markert

Wolfgang Fischer

Martin Schiegg

Sebastian Vogel

Christoph Schorn

Juergen Schirmer

Ernst Kloppenburg

Michael Hanselmann

Jaroslaw Topp

Leonardo Luiz Ecco

Armin Runge

Norbert Wehn

Nico Bannow

Felix Streichert

Taha Ibrahim Ibrahim Soliman

Chirag Sudarshan

Christian Weis

Axel Aue

Tobias Lang

Matthias Schreiber

Holger Ulmer

Jo Pletinckx

Ulrich Schulmeister

Christian Fleck

Rene Diener

Michael Saetzler

Anne Von Vietinghoff

A method for operating a hardware platform for the inference calculation of a layered neural network. In the method: a first portion of input data which are required for the inference calculation of a first layer of the neural network and redundancy information relating to the input data are read in from an external working memory into an internal working memory of the computing unit; the integrity of the input data is checked based on the redundancy information; in response to the input data here being identified as error-free, the computing unit carries out at least part of the first-layer inference calculation for the input data to obtain a work result; redundancy information for the work result is determined, based which the integrity of the work result can be verified; the work result and the redundancy information are written to the external working memory.

Inference calculation for neural networks with protection against memory errors

A memory device comprising a plurality of memory cells situated in a first cell field, multiple first bit lines, each respectively connected to multiple memory cells of the first cell field to enable access to the memory cells via the bit line, and multiple sense amplifier pairs which respectively comprise a first and a second sense amplifier. Each first bit line is assigned to a sense amplifier pair, each first bit line being connected to a respective first semiconductor switch element, through which the bit line is electroconductively connectible to and insulatable from the first sense amplifier of the sense amplifier pair, to which the bit line is assigned. Each first bit line is connected to a respective second semiconductor switch element, through which the bit line is electroconductively connectible to and insulatable from the second sense amplifier of the sense amplifier pair, to which the bit line is assigned.

Memory device and method for performing consecutive memory accesses

A memory device comprising a cell field having memory cells, N bit lines, which are respectively connected to at least one of the memory cells of the cell field, N being a whole number greater than one, N sense amplifiers; a bit shift circuit, which has S switch element rows, S being a whole number greater than one and a row number in the range from zero to S−1 being assignable to each switch element row. Each switch element row includes at least one semiconductor switch element connected to one of the bit lines and one of the sense amplifiers. Switch elements of each row connect all bit lines, whose bit line number is smaller than or equal to N minus the row number, to sense amplifiers, so that the respective sense amplifier number is equal to the respective bit line number plus the row number.

Memory device and method for shifting memory values

A method for operating a calculation system including a neural network, in particular a convolutional neural network, the calculation system including a processing unit for the sequential calculation of the neural network and a memory external thereto for buffering intermediate results of the calculations in the processing unit, including: incrementally calculating data sections, which each represent a group of intermediate results, with the aid of a neural network; lossy compression of one or multiple of the data sections to obtain compressed intermediate results; and transmitting the compressed intermediate results to the external memory.

Method and device for operating a neural network in a memory-efficient manner

A hardware architecture for an artificial neural network ANN. The ANN includes a consecutive series made up of an input layer, multiple processing layers, and an output layer. Each layer maps a set of input variables onto a set of output variables, and output variables of the input layer and of each processing layer are input variables of the particular layer that follows in the series. The hardware architecture includes a plurality of processing units. The implementation of each layer is split among at least two of the processing units, and at least one resettable switch-off device is provided via which at least one processing unit is selectively deactivatable, independently of the input variables supplied to it, in such a way that at least one further processing unit remains activated in all layers whose implementation is contributed to by this processing unit.

Selective deactivation of processing units for artificial neural networks

A model calculating unit for calculating a neural layer of a multilayer perceptron model having a hardwired processor core developed in hardware for calculating a definitely specified computing algorithm in coupled functional blocks. The processor core is designed to calculate, as a function of one or multiple input variables of an input variable vector, of a weighting matrix having weighting factors and an offset value specified for each neuron, an output variable for each neuron for a neural layer of a multilayer perceptron model having a number of neurons, a sum of the values of the input variables weighted by the weighting factor, determined by the neuron and the input variable, and the offset value specified for the neuron being calculated for each neuron and the result being transformed using an activation function in order to obtain the output variable for the neuron.

Model calculating unit and control unit for calculating a neural layer of a multilayer perceptron model

A model calculation unit for calculating an RBF model is described, including a hard-wired processor core designed as hardware for calculating a fixedly predefined processing algorithm in coupled functional blocks, the processor core being designed to calculate an output variable for an RBF model as a function of one or multiple input variable(s) of nodes V[j,k], of length scales (L[j,k]), of weighting parameters p3[j,k] predefined for each node, the output variable being formed as a sum of a value calculated for each node V[j,k], the value resulting from a product of a weighting parameter p3[j,k] assigned to the particular node V[j,k], and a result of an exponential function of a value resulting from the input variable vector as a function of a square distance of the particular node (V[j,k]), weighted by the length scales (L[j,k]), the length scales (L[j,k]) being provided separately for each of the nodes as local length scales.

Model calculation unit and control unit for calculating an RBF model

A hardware-implemented multi-layer perceptron model calculation unit includes: a processor core to calculate output quantities of a neuron layer based on input quantities of an input vector; a memory that has, for each neuron layer, a respective configuration segment for storing configuration parameters and a respective data storage segment for storing the input quantities of the input vector and the one or more output quantities; and a DMA unit to successively instruct the processor core to: calculate respective neuron layers based on the configuration parameters of each configuration segment, calculate input quantities of the input vector defined thereby, and store respectively resulting output quantities in a data storage segment defined by the corresponding configuration parameters, the configuration parameters of configuration segments successively taken into account indicating a data storage region for the resulting output quantities corresponding to the data storage region for the input quantities for a subsequent neuron layer.

Model calculation unit and control device for calculating a multi-layer perceptron model

A method for verifying a calculation of a neuron value of multiple neurons of a neural network, including: carrying out or triggering a calculation of neuron functions of the multiple neurons, in each case to obtain a neuron value, the neuron functions being determined by individual weightings for each neuron input; calculating a first comparison value as the sum of the neuron values of the multiple neurons; carrying out or triggering a control calculation with one or multiple control neuron functions and with all neuron inputs of the multiple neurons, to obtain a second comparison value as a function of the neuron inputs of the multiple neurons and of the sum of the weightings of the multiple neurons assigned to the respective neuron input; and recognizing an error as a function of the first comparison value and of the second comparison value.

Method and device for verifying a neuron function in a neural network

A model calculating unit for the selective calculation of an RBF model or of a neural layer of a multilayer perceptron model having a hardwired processor core designed in hardware for calculating a fixedly specified computing algorithm in coupled function blocks. The processor core is designed to calculate an output variable for an RBF model as a function of one or multiple input variables of an input variable vector, of supporting points, of length scales, of parameters specified for each supporting point, the processor core furthermore being designed to calculate an output variable for each neuron for the neural layer of the multilayer perceptron model having a number of neurons as a function of the one or the multiple input variables of the input variable vector, of a weighting matrix having weighting factors and an offset value specified for each neuron.

Model calculating unit and control unit for selectively calculating an RBF model, a gaussian process model and an MLP model

A model calculation unit for calculating a multilayer perceptron model, the model calculation unit being designed in hardware and being hardwired, including: a process or core; a memory; a DMA unit, which is designed to successively instruct the processor core to calculate a neuron layer, in each case based on input variables of an assigned input variable vector and to store the respectively resulting output variables of an output variable vector in an assigned data memory section, the data memory section for the input variable vector assigned to at least one of the neuron layers at least partially including in each case the data memory sections of at least two of the output variable vectors of two different neuron layers.

Model calculation unit and control unit for calculating a multilayer perceptron model with feedforward and feedback

A model calculation unit for calculating a gradient with respect to a certain input variable of input variables of a predefined input variable vector for an RBF model with the aid of a hard-wired processor core designed as hardware for calculating a fixedly predefined processing algorithm in coupled functional blocks, the processor core being designed to calculate the gradient with respect to the certain input variable for an RBF model as a function of one or multiple input variable(s) of the input variable vector of an input dimension, of a number of nodes, of length scales predefined for each node and each input dimension, and of parameters of the RBF function predefined for each node.

Model calculation unit and control unit for calculating a partial derivative of an RBF model

A method for calculating a neuron layer of a multi-layer perceptron model that includes a permanently hardwired processor core configured in hardware for calculating a permanently predefined processing algorithm in coupled functional blocks, a neuron of a neuron layer of the perceptron model being calculated with the aid of an activation function, the activation function corresponding to a simplified sigmoid function and to a simplified tan h function, the activation function being formed by zero-point mirroring of the negative definition range of the exponential function.

Method for calculating a neuron layer of a multi-layer perceptron model with simplified activation function

A model calculation unit for calculating a data-based function model in a control unit is provided, the model calculation unit having a processor core which includes: a multiplication unit for carrying out a multiplication on the hardware side; an addition unit for carrying out an addition on the hardware side; an exponential function unit for calculating an exponential function on the hardware side; a memory in the form of a configuration register for storing hyperparameters and node data of the data-based function model to be calculated; and a logic circuit for controlling, on the hardware side, the calculation sequence in the multiplication unit, the addition unit, the exponential function unit and the memory in order to ascertain the data-based function model.

Model calculation unit, control unit and method for calibrating a data-based function model

A method for carrying out a calculation of a data-based function model in a control unit having a computing unit and a separate model calculation unit having a computing core, including: loading a first part of the configuration data, which contain hyperparameters of the data-based function model and a first part of supporting point data having multiple supporting points, into the model calculation unit; starting a calculation in the computing core of the model calculation unit, to obtain a model value at a predefined test point; and transferring a second part of the configuration data, which contain a second part of the supporting point data having multiple supporting points, into the model calculation unit, prior to the completion of the calculation in the computing core of the model calculation unit.

Growing community of inventors

Pforzheim, Germany

Andre Guntoro