International Business Machines Corporation

Globalfoundries Inc.

Stefan Abel

Jean Fompeyrine

Veeresh Vidyadhar Deshpande

Lukas Czornomaz

Bert Jan Offrein

Antonio La Porta

Chiara Marchiori

Paul F Seidler

Pascal Stark

Bernd W Gotsmann

Fabian Menges

Charles Caër

Lorenz K Muller

Lukas Czomomaz

Youri Popoff

Oleg Gluschenkov

Vijay Narayanan

Hemanth Jagannathan

Abu Sebastian

Martin Michael Frank

Eduard Albert Cartier

Angeliki Pantazi

Patrick Ruch

Walter Heinrich Riess

Folkert Horst

Kam-Leung Lee

Stanislaw Andrzej Wozniak

Marc Seifried

Johannes Gooth

Utz Herwig Hahn

Valeria Bragaglia

Karl Felix Sebastian Eltes

Mario El Kazzi

A method of operating a neural network. The input layer of the network may have n input nodes connected to output nodes via a hidden layer. The hidden layer may include m hidden nodes. The n input nodes may connect to a subset of k nodes of the m hidden nodes via respective synaptic connections, to which training weights are associated, which form an n×k input matrix W, whereas a subset of m−k nodes of the hidden layer are not connected by any node of the input layer. Running the network may include performing a first matrix vector multiplication between the input matrix Wand a vector of values obtained in output of the input nodes and a second matrix vector multiplication between a fixed matrix Wof fixed weights and a vector of values obtained in output of the m nodes of the hidden layer.

Fixed, random, recurrent matrices for increased dimensionality in neural networks

A method and computer program product for obtaining values are run using a neural network according to a machine learning algorithm. One embodiment may comprise accessing one or more datafiles of input data, where the input data is representable in a d-dimensional space, with d>1. The method may explore N distinct paths of the input data in the d-dimensional space, where N≥1, and collects data along the N distinct paths explored to respectively form N sequences of M objects each, with M≥2. For one or more sequences of the N sequences formed, values obtained from the M objects of each sequence may be coupled into one or more input nodes of a neural network, which is then run according to the machine learning algorithm to obtain L output values from, L≥1.

Extracting sequences from d-dimensional input data for sequential processing with neural networks

An optical synapse comprises a memristive device for non-volatile storage of a synaptic weight dependent on resistance of the device, and an optical modulator for volatile modulation of optical transmission in a waveguide. The memristive device and optical modulator are connected in control circuitry which is operable, in a write mode, to supply a programming signal to the memristive device to program the synaptic weight and, in a read mode, to supply an electrical signal, dependent on the synaptic weight, to the optical modulator whereby the optical transmission is controlled in a volatile manner in dependence on programmed synaptic weight.

Optical synapses

An integrated optical circuit for an optical neural network is provided. The integrated optical circuit is configured to process a phase-encoded optical input signal and to provide a phase-encoded output signal depending on the phase-encoded optical input signal. The phase-encoded output signal emulates a synapse functionality with respect to the phase-encoded optical input signal. A related method and a related design structure are further provided.

Optical synapse

An electrochemical device includes an electrochemical cell and an electric circuit. The electrochemical cell comprises a first solid component and a second solid component. The two solid components comprise same chemical elements but have different concentrations of at least one type of the chemical elements. A solid electrolyte is arranged between the two solid components. The solid electrolyte is a dielectric material. The electric circuit is connected to the electrochemical cell. The electrochemical cell may be operated according to a redox process, so as to exchange chemical elements of the at least one type between the first solid component and the second solid component and thereby change an electrical conductance of each of the two solid components.

Electrochemical device of variable electrical conductance

Embodiment of the invention are directed to transmitting signals between neurons of a hardware-implemented, spiking neural network (or SNN). The network includes neuronal connections, each including a synaptic unit connecting a pre-synaptic neuron to a post-synaptic neuron. Spikes received from the pre-synaptic neuron of said each neuronal connection are first modulated, in frequency, based on a synaptic weight stored on said each synaptic unit, to generate post-synaptic spikes, such that a first number of spikes received from the pre-synaptic neuron are translated into a second number of post-synaptic spikes. At least some of the spikes received from the pre-synaptic neuron may, each, be translated into a train of two or more post-synaptic spikes. The post-synaptic spikes generated are subsequently transmitted to the post-synaptic neuron of said each neuronal connection. The novel approach makes it possible to obtain a higher dynamic range in the synapse output.

Spiking synaptic elements for spiking neural networks

An integrated optical circuit for an optical neural network is provided. The optical circuit is configured to process a plurality of phase-encoded optical input signals and to provide a phase-encoded optical output signal depending on the phase-encoded optical input signals. The phase-encoded optical output signal emulates a neuron functionality with respect to the plurality of phase-encoded optical input signals. Such an embodied optical circuit uses the phase to encode information in the optical domain. A related method and a related design structure are further provided.

Optical neuron

A method for hardware-implemented training of a feedforward artificial neural network is provided. The method comprises: generating a first output signal by processing an input signal with the network, wherein a cost quantity assumes a first cost value; measuring the first cost value; defining a group of at least one synaptic weight of the network for variation; varying each weight of the group by a predefined weight difference; after the variation, generating a second output signal from the input signal to measure a second cost value; comparing the first and second cost values; and determining, based on the comparison, a desired weight change for each weight of the group such that the cost function does not increase if the respective desired weight changes are added to the weights of the group. The desired weight change is based on the weight difference times &#x2212;1, 0, or +1.

Hardware-implemented training of an artificial neural network

Embodiments are directed to the fabrication of an electro-optical device. The device comprises the forming of an active region with a stack of III-V semiconductor gain materials stacked along a stacking direction z. The active region may be formed as a slab having several lateral surface portions, each extending parallel to the stacking direction z. The device further comprises selectively re-growing two paired elements, which include: a pair of doped layers of III-V semiconductor materials (an n-doped layer and a p-doped layer); and a pair of lateral waveguide cores. The two paired elements may be laterally arranged, two-by-two, on opposite sides of the slab. The elements distinctly adjoin respective ones of the lateral surface portions of the slab, so as for these elements to be separated from each other by the slab. The disclosure may be further directed to related silicon photonics devices.

Method of forming an electro-optical device with lateral current injection regions

An artificial cochlea device for processing audio signals by electro-mechanical amplitude changes may be provided. The device comprises a micro-electro-mechanical system (MEMS) microphone comprising a membrane, an electro-mechanical feedback loop embedded in the MEMS microphone. Thereby, the electro-mechanical feedback loop comprises a piezo-electric actuator acting on the MEMS microphone by influencing the membrane's way to swing, such that an electro-mechanical amplitude change of an output signal is achieved.

Artificial cochlea

Aspects of the invention are directed to an electro-optical device having a layer structure including a substrate, an electrically insulating layer on top of the substrate, a bonding layer on top of the electrically insulating layer, a Pockels layer on top of the bonding layer, a waveguide core on top of the Pockels layer, and a cladding layer cladding both the waveguide core and the Pockels layer, the latter coated by the cladding layer. The Pockels layer is a layer of a crystalline first material having a Pockels coefficient between 10 pm/V and 10 000 pm/V. The waveguide core includes a second material, which can be crystalline. The device can be adapted to optically couple radiation into and/or from the waveguide core. Each of the first material and the second material has a larger refractive index than the electrically insulating layer and the cladding layer for said radiation.

Integrated waveguide structure with pockels layer having a selected crystal orientation

A tunable resistive element, comprising a first terminal, a second terminal, a dielectric layer and an intercalation layer. The dielectric layer and the intercalation layer is arranged between the first terminal and the second terminal. The dielectric layer is configured to form conductive filaments of oxygen vacancies on application of an electric field. The intercalation layer is configured to undergo a topotactic transition comprising an oxygen intercalation in combination with a change in the resistivity of the intercalation layer. A related memory device and a related neuromorphic network comprise resistive memory elements as memory cells and synapses respectively and a corresponding design structure.

Tunable resistive element

An integrated optical transmission element may be provided. The integrated optical transmission element includes an optical cavity including an input port and an output port, and photorefractive material within the optical cavity. A transmission of light from the input port to the output port is persistently changeable by an optical control signal provided to the photorefractive material, the optical control signal being configured to change a refractive index.

Integrated optical transmission element

A method for converting a dielectric material including a type IV transition metal into a crystalline material that includes forming a predominantly non-crystalline dielectric material including the type IV transition metal on a supporting substrate as a component of an electrical device having a scale of microscale or less; and converting the predominantly non-crystalline dielectric material including the type IV transition metal to a crystalline crystal structure by exposure to energy for durations of less than 100 milliseconds and, in some instances, less than 10 microseconds. The resultant material is fully or partially crystallized and contains a metastable ferroelectric phase such as the polar orthorhombic phase of space group Pca2or Pmn2. During the conversion to the crystalline crystal structure, adjacently positioned components of the electrical devices are not damaged.

Fast recrystallization of hafnium or zirconium based oxides in insulator-metal structures

Demodulating a modulated signal. A method may include receiving a modulated signal, wherein the modulated signal is a signal modulated according to a modulation function varying faster than the signal. The modulation function is a function of the signal. The modulated signal received is demodulated with an artificial neural network system, or ANN system, which is trained to identify bit values from signal patterns as caused by the modulation function, by identifying bit values from patterns of the modulated signal received. Related modulation and demodulation systems are disclosed.

Growing community of inventors

Zurich, Switzerland

Stefan Abel