Commissariat &#xc0; L'&#xc9;nergie Atomique Et Aux &#xc9;nergies Alternatives

Marc Duranton

Jean-Marc Philippe

Olivier Bichler

Alexandre Carbon

Michel Paindavoine

Vincent Lorrain

Johannes Christian Thiele

Rodolphe Heliot

Stéphane Chevobbe

Stephane Chevobbe

Anthony Kolar

Antoine Joubert

Hong-Phuc Trinh

A computer-implemented method for coding a digital signal intended to be processed by a digital computing system includes the steps of: receiving a sample of the digital signal quantized on a number Nof bits, decomposing the sample into a plurality of binary words of parameterizable bit size N, coding the sample through a plurality of pairs of values, each pair comprising one of the binary words and an address corresponding to the position of the binary word in the sample, transmitting the pairs of values to an integration unit in order to carry out a MAC operation between the sample and a weighting coefficient.

Method and device for binary coding of signals in order to implement digital MAC operations with dynamic precision

A circuit comprises a series of calculating blocks that can each implement a group of neurons; a transformation block that is linked to the calculating blocks by a communication means and that can be linked at the input of the circuit to an external data bus, the transformation block transforming the format of the input data and transmitting the data to said calculating blocks by means of K independent communication channels, an input data word being cut up into sub-words such that the sub-words are transmitted over multiple successive communication cycles, one sub-word being transmitted per communication cycle over a communication channel dedicated to the word such that the N channels can transmit K words in parallel.

Electronic circuit, particularly for the implementation of neural networks with multiple levels of precision

A neuron circuit is capable of producing a weighted sum of digitized input signals and applying an activation function to the weighted sum so as to produce a digitized activation signal as output. The circuit includes at least: one multiplier multiplying each input signal (xto x) with a weight value (wto w), one accumulator accumulating the results of the multiplier so as to produce the weighted sum, and one activation unit executing the activation function. The activation unit comprises at least one shift unit and at least one saturation unit capable of approximating a non-linear activation function. The result of the approximated activation function is obtained by one or more arithmetic shifts applied to the weighted sum.

Optimized neuron circuit, and architecture and method for executing neural networks

A processor having a SIMD architecture, including an array of elementary processors, each elementary processor being associated with an elementary memory cell, a central controller connected to the elementary processors by an instruction bus and a status bus. The central controller transmits a sequence of instructions in a loop, each instruction including a calculation flow indicator. Each elementary processor has an instruction filter that makes it possible to reject or take into account an instruction depending on the identifier it contains. This operating mode makes it possible to emulate a MIMD processor on a SIMD architecture.

MIMD processor emulated on SIMD architecture

An implementation of neural networks on silicon for the processing of various signals comprises multidimensional signals such as images. The efficient implementation on silicon of a complete processing chain for the signal via the approach using neural networks is provided. The circuit comprises at least: a series of neuro-blocks grouped together in branches composed of a group of neuro-blocks and a broadcasting bus, the neuro-blocks connected to the broadcasting bus; a routing unit connected to the broadcasting bus of the branches, carrying out the routing and broadcasting of data to and from the branches; a transformation module connected to the routing unit via an internal bus and designed to be connected at the input of the circuit to an external databus, the module carrying out the transformation of input data into serial coded data. The processing operations internal to the circuit are carried out according to a serial communications protocol.

Electronic circuit, in particular capable of implementing a neural network, and neural system

A signal processing module comprises at least one operational unit incorporating computation units, input and output interfaces able to be linked to a bus and a memory storing data destined for the computation units, the memory being organized so that each data word is stored column-wise over several addresses according to an order dependent on the application, a column having a width of one bit, the words being transferred in series to the computation units.

Signal processing module, especially for a neural network and a neuronal circuit

A method implemented by a processor for coding a real signal, for example an image signal, into a quantized signal, comprises the following steps applied to each real sample of the real signal: converting the real sample into a digital representation, selecting, in the fractional part of the number, a predetermined number N of most significant non-zero bits, for each non-zero significant bit i selected, i varying from 1 to N, determining its distance Pwith respect to the neighboring selected non-zero significant bit of higher rank or, for the first non-zero significant bit selected, with respect to the decimal point, deducting from the distance Pthe minimum value of distance between two non-zero bits, coding the modified distance Pon a predetermined number Mof bits.

Method of coding a real signal into a quantized signal

A communication system coupled to a data-acquisition circuit and to a data-processing circuit is provided, including at least one shift register, an addressing circuit and a multiplexer. The shift register includes a serial input for inputting and storing data in series, a serial output for outputting data in series, and parallel outputs for outputting data stored in the shift register in parallel. The addressing circuit is coupled to the shift register in order to identify the positions of stored data, and the multiplexer is coupled to the parallel outputs of the shift register in order to output the stored data to the data-processing circuit in series. Methods for communication between a data-acquisition circuit and a data-processing circuit are also provided.

System and method for communication between a data-acquisition circuit and a data-processing circuit

A three-dimensional scene sensor comprises: a deformable optical system modifying focal distance by control signal, optics imaging the scene by analog image sensor for depths corresponding to distances; the image sensor comprising a matrix of pixels grouped into sub-matrices of macro-pixels being a sub-assembly of pixels, each macro-pixel operating independently for acquisition and reading of data; a matrix of elementary processors, each macro-pixel directly connected to a dedicated processor wherein pixel data for the macro-pixel are transmitted and processed by the processor, each processor carries out, for each pixel, local processing operations calculating depth information for macro-pixel, the processors operating in parallel and independently such that the depth information is processed and calculated in parallel over all macro-pixels of the image sensor, the processors connected to at least one processing unit allowing calculations using high-level input data, calculated starting from the pixel data directly produced by the image sensor.

Integrated three-dimensional vision sensor

A neuromorphic data processing device comprising a plurality of spiking neurons, with each of these neurons comprising: an integrator designed to receive successive analog pulses each having a certain value, and accumulate the values of the pulses received in a recorded value, referred to as accumulation value, and a discharger designed to emit a pulse, referred to as discharge pulse, according to the accumulation value, and a silicon support having two surfaces, the neurons being carried out on at least one of the two surfaces, the integrator of each neuron comprising a metal via of the TSV type between the two surfaces of the silicon support, the metal via of the TSV type forming a capacitor with the silicon support and having an electric potential forming the accumulation value wherein the values of the pulses received are accumulated and according to which the discharge pulse is emitted.

Growing community of inventors

Orsay, France

Marc Duranton