Imec Vzw

Imec

Katholieke Universiteit Leuven

Katholieke Universiteit Leuven, Ku Leuven R&d

Samsung Electronics Co., Ltd.

Stichting Imec Nederland

Vrije Universiteit Brussel

Katholieke Universiteit Leuven, K.u. Leuven Research & Development

Praveen Raghavan

Francky Catthoor

Julien Ryckaert

Juergen Boemmels

Odysseas Zografos

Raf Appeltans

Dimitrios Rodopoulos

Frederik Naessens

Robert Fasthuber

Andy Lambrechts

Stefan Cosemans

Vassilios Constantinos Gerousis

Murali Jayapala

Trong Huynh Bao

Pieter Weckx

Steven Demuynck

Ryan Ryoung Han Kim

Davide Francesco Crotti

Sofie Pollin

Daniele Garbin

Komalan Manu Perumkunnil

Wim Van Thillo

Jos Huisken

Jose Ignacio Gomez

Christian Tenllado

Peter Debacker

Matthias Hartmann

Eduardo Lopez Estraviz

David Novo Bruna

Angeliki Krithikakou

Absar Javed

Youssef Drissi

Mohit Dandekar

Stephane Lariviere

Darko Trivkovic

Evangelos Bempelis

Elena Hammari

Per Gunnar Kjeldsberg

Lieven Hollevoet

A method and hardware system for mapping an input map of a convolutional neural network layer to an output map are disclosed. An array of processing elements are interconnected to support unidirectional dataflows through the array along at least three different spatial directions. Each processing element is adapted to combine values of dataflows along different spatial directions into a new value for at least one of the supported dataflows. For each data entry in the output map, a plurality of products from pairs of weights of a selected convolution kernel and selected data entries in the input map is provided and arranged into a plurality of associated partial sums. Products associated with a same partial sum are accumulated on the array and accumulated on the array into at least one data entry in the output map.

Convolution engine for neural networks

A control plane for controlling transfer of data to a data plane is disclosed. In one aspect, the control plane comprises memory cells for storing a digitally coded parameter value and having a data input electrode, a data output electrode and a control electrode, n data input terminals that receive a data input value and apply it to the data input electrode of an associated memory cell, and n data output terminals coupled to a data output electrode of an associated memory cell. The control plane further comprise a first delay line having delay elements and arranged for receiving a stream of control bit values, and a second delay line having delay elements and arranged for receiving a signal for enabling the control bit values in the first delay line, wherein data is transferred in a controlled and synchronized fashion to an output electrode.

Control plane organization for flexible digital data plane

The disclosed technology generally relates to magnetoresistive devices, and more particularly to a magnetic tunnel junction (MTJ) device formed in an interconnection structure, and to a method of integrating the (MTJ) device in the interconnection structure. According to an aspect, a device includes a first interconnection level including a first dielectric layer and a first set of conductive paths arranged in the first dielectric layer, a second interconnection level arranged on the first connection level and including a second dielectric layer and a second set of conductive paths arranged in the second dielectric layer, and a third interconnection level arranged on the second interconnection level and including a third dielectric layer and a third set of conductive paths arranged in the third dielectric layer. The device additionally includes a magnetic tunnel junction (MTJ) device including a bottom layer, a top layer and an MTJ structure arranged between the bottom layer and the top layer, wherein the bottom layer is connected to a bottom layer contact portion of the first set of conductive paths and the top layer is connected to a top layer contact portion of the second or third set of conductive paths. The device further includes a multi-level via extending through the second dielectric layer and the third dielectric layer, between a first via contact portion of the first set of conductive paths and a second via contact portion of the third set of conductive paths, wherein a height of the MTJ device corresponds to, or-is less than, a height of the multi-level via, e.g., wherein the height of the MTJ device corresponds to or is less than a height of the second interconnection level.

Magnetoresistive device and method of fabricating same

An integrated circuit (IC) power distribution network is disclosed. In one aspect, the IC includes a stack of layers formed on a substrate. The IC includes standard cells with parallel gate structures oriented in a direction y. Each cell includes an internal power pin for supplying a reference voltage to the cell. The stack includes metal layers in which lines are formed to route signals between cells. The lines in each metal layer have a preferred orientation that is orthogonal to that of the lines in an adjacent metal layer. A first layer is the lowest metal layer that has y as a preferred orientation while also providing routing resources for signal routing between the cells. A second layer is the nearest metal layer above this first layer. The IC includes a power distribution network for delivering the reference voltage to the power pin.

Integrated circuit power distribution network

A semiconductor device is disclosed that includes a substrate and at least a first, second, third, and fourth vertical transistor supported by the substrate. Each transistor comprises a vertical channel, a polarity gate electrode forming a polarity gate adapted to act on a first portion of the channel to affect a polarity of the channel, and a control gate electrode forming a control gate adapted to act on a second portion of the channel to control the electrical conductivity of the channel. The polarity gate electrode and the control gate electrode of each one of the transistors extend laterally from their respective gate and in mutually opposite directions, and the transistors are laterally spaced from each other and arranged such that the control gate electrodes of the first and third transistor face each other and the control gate electrodes of the second and fourth transistor face each other.

Double-gate vertical transistor semiconductor device

In an aspect of the disclosed technology, a SRAM device includes a first stack of transistors and a second stack of transistors arranged on a substrate. Each of the first and second stacks includes a pull-up transistor, a pull-down transistor and a pass transistor, where each of the transistors includes a horizontally extending channel. In each of the first and second stacks, the pull-up transistor and the pull-down transistor have a common gate electrode extending vertically therebetween, and the pass transistor has a gate electrode separated from the common gate electrode. A source/drain of each of the pull-up transistor and the pull-down transistor and a source/drain of the pass transistor included in one of the first stack and the second stack are electrically interconnected with the common gate electrode of the pull-up transistor and the pull-down transistor included in the other of the first stack and the second stack.

Static random access memory device having interconnected stacks of  transistors

A method of forming conductive paths and vias is disclosed. In one aspect, patterns of a hard mask layer are transferred into a dielectric layer by etching to form trenches. The trenches define locations for conductive paths of an upper metallization level. At least one trench is interrupted in a longitudinal direction by a block portion of the hard mask layer, the block portion defining the tip-to-tip location of a pair of the conductive paths to be formed. The trenches extend partially through the dielectric layer in regions exposed by the hard mask layer, thereby deepening first and the second holes to extend completely through the dielectric layer. After removing the hard mask layer, the deepened first and second holes and the trenches are filled with a conductive material to form the conductive paths in the trenches and to form the vias in the deepened first and second holes.

Methods for forming conductive paths and vias

Three transistor two junction magnetoresistive random-access memory (MRAM) bit cells provided. An example MRAM bit cell includes a first magnetic tunnel junction, MTJ, connected to a first bit line. The MRAM bit cell also includes a second MTJ connected to a second bit line. In addition, the MRAM bit cell includes a first transistor connected to the first MTJ and to a ground conductor. The MRAM bit cell further includes a second transistor connected to the second MTJ and to the ground conductor. Additionally, the MRAM bit cell includes a third transistor connected to the first transistor and to the second transistor.

Three transistor two junction magnetoresistive random-access memory (MRAM) bit cell

The disclosure relates to a data communication network connecting a plurality of computation clusters. The data communication network is arranged for receiving via N data input ports, N&#x3e;1, input signals from first clusters of the plurality and for outputting output signals to second clusters of the plurality via M data output ports, M&#x3e;1. The communication network includes a segmented bus network for interconnecting clusters of the plurality and a controller arranged for concurrently activating up to P parallel data busses of the segmented bus network, thereby forming bidirectional parallel interconnections between P of the N inputs, P&#x3c;N, and P of the M outputs, P&#x3c;M, via paths of connected and activated segments of the segmented bus network. The segments are linked by segmentation switches. The N data input ports and the M data output ports are connected via stubs to a subset of the segmentation switches.

Inter-cluster data communication network for a dynamic shared communication platform

A semiconductor circuit comprises a Front End of Line (FEOL) comprising a plurality of transistors, each of which having a source region, a drain region and a gate region arranged between the source region and the drain region and comprising a gate electrode. The semiconductor circuit also comprises a buried interconnect that is arranged in the FEOL and electrically connected to the gate region from below through a bottom contact portion of the gate electrode. By using a buried interconnect the routing of the circuit may be facilitated.

Buried interconnect for semiconductor circuits

The present disclosure relates to low-layer memory for a computing platform. An example embodiment includes a memory hierarchy being directly connectable to a processor. The memory hierarchy includes at least a level, referred to as L, memory structure comprising a non-volatile memory unit as Ldata memory and a buffer structure (L-VWB). The buffer structure includes a plurality of interconnected wide registers with an asymmetric organization, wider towards the non-volatile memory unit than towards a data path connectable to the processor. The buffer structure and the non-volatile memory unit are arranged for being directly connectable to a processor so that data words can be read directly from either of the Ldata memory and the buffer structure (L-VWB) by the processor.

Low-layer memory for a computing platform

The present system and method relate to a system for performing a multiplication. The system is arranged for receiving a first data value, and comprises means for calculating at run time a set of instructions for performing a multiplication using the first data value, storage means for storing the set of instructions calculated at run time, multiplication means arranged for receiving a second data value and at least one instruction from the stored set of instructions and arranged for performing multiplication of the first and the second data values using the at least one instruction.

System and method for implementing a multiplication

Methods are disclosed for system scenario-based design for an embedded platform whereon a dynamic application is implemented. The application meets at least one guaranteed constraint. Temporal correlations are assumed in the behavior of internal data variables used in the application, with the internal data variables representing parameters used for executing a portion of the application. An example method includes determining a distribution over time of an N-dimensional cost function, with N an integer number N&#x2267;1, corresponding to the implementation on the platform for a set of combinations of the internal data variables. The method also includes partitioning an N-dimensional cost space in at least two bounded regions, each bounded region containing cost combinations corresponding to combinations of values of the internal data variables of the set that have similar cost and frequency of occurrence, whereby one bounded region is provided for rarely occurring cost combinations.

Method for system scenario based design of dynamic embedded systems

A digital front-end circuit is disclosed. In one aspect, the circuit includes a filtering block for filtering received data. The filtering block has a first filter branch for filtering the received data in a first frequency band and a second filter branch for filtering the received data in a selected second frequency band. The second filter branch is in parallel with the first filter branch, is programmable and includes a block for resampling the received data. The front-end circuit also includes a circuit for performing synchronization and spectrum sensing on the received data, which is in connection with the output of the filtering block. The front-end circuit also includes a controller block for controlling the filtering block and the synchronization circuit.

Digital front-end circuit and method for using the same

Disclosed is a method for selecting a design option for a Viterbi decoder model. In some embodiments, the method includes deriving a set of design options for a Viterbi decoder model by differentiating at least one design parameter, where the at least one design parameter comprises at least a first value for a look-ahead parameter. The method further includes performing an evaluation of each design option in the set of design options in a multi-dimensional design space and, based on the evaluation of each design option, selecting a design option in the set of design options that (i) satisfies a predetermined energy efficiency constraint and (ii) yields at least a second value for the look-ahead parameter, wherein the second value is greater than the first value and satisfies a predetermined area budget.

Methods for viterbi decoder implementation

Growing community of inventors

Leefdaal, Belgium

Praveen Raghavan