Intel Corporation

Enric Herrero Abellanas

Pedro Lopez

Marc Lupon

Ayose J Falcon

Frederico C Pratas

Fernando Latorre

Xavier Vera

Javier Carretero Casado

Tanausu Ramirez

Nicholas Axelos

Daniel Sanchez

Georgios Tournavitis

Matteo Monchiero

Pedro López

Ayose J Falcón

Marc Lupón

An apparatus and method are described for distributed and cooperative computation in artificial neural networks. For example, one embodiment of an apparatus comprises: an input/output (I/O) interface; a plurality of processing units communicatively coupled to the I/O interface to receive data for input neurons and synaptic weights associated with each of the input neurons, each of the plurality of processing units to process at least a portion of the data for the input neurons and synaptic weights to generate partial results; and an interconnect communicatively coupling the plurality of processing units, each of the processing units to share the partial results with one or more other processing units over the interconnect, the other processing units using the partial results to generate additional partial results or final results. The processing units may share data including input neurons and weights over the shared input bus.

Method and apparatus for distributed and cooperative computation in artificial neural networks

A processing device includes a processor core and a number of calculation modules that each is configurable to perform any one of operations for a convolutional neuron network system. A first set of the calculation modules are configured to perform convolution operations, a second set of the calculation modules are reconfigured to perform averaging operations, and a third set of the calculation modules are reconfigured to perform dot product operations.

Reconfigurable processing unit

Systems and methods for performing convolution operations. An example processing system comprises: a processing core; and a convolver unit to apply a convolution filter to a plurality of input data elements represented by a two-dimensional array, the convolver unit comprising a plurality of multipliers coupled to two or more sets of latches, wherein each set of latches is to store a plurality of data elements of a respective one-dimensional section of the two-dimensional array.

Processing device for performing convolution operations

A storage device and method are described for performing convolution operations. For example, one embodiment of an apparatus to perform convolution operations comprises a plurality of processing units to execute convolution operations on input data and partial results; a unified scratchpad memory comprising a plurality of memory banks communicatively coupled to the plurality of processing units through a plurality of read/write ports, each of the plurality of memory banks partitioned to store both the input data and partial results; a control unit to allocate the input data and partial results to the memory banks to ensure a minimum quality of service in accordance with the specified number of read/write ports and the specified convolution operation to be performed.

Storage device and method for performing convolution operations

An apparatus, system, and method for controlling traffic on an on-chip network. Embodiments of the method comprise injecting a packet at a first rate into the on-chip network by a first node coupled to the on-chip network, receiving the packet at a second node coupled to the on-chip network, modifying a bit in the packet by the second node in response to determining that a rate at which packets are injected into the on-chip network should change, returning the packet with the bit modified to the first node by the second node, and changing the first rate by the first node in response to detecting that the bit in the packet was modified.

Traffic control on an on-chip network

A system, processor and method to reduce the overall detectable unrecoverable FIT rate of a cache by reducing the residency time of dirty lines in a cache. This is accomplished through selectively choosing different replacement policies during execution based on the DUE FIT target of the system. System performance and power is minimally affected while effectively reducing the DUE FIT rate.

Increased error correction for cache memories through adaptive replacement policies

A method of determining vulnerability of a cache memory includes associating a first counter with a cache element and periodically incrementing the first counter. When a read or other access that consumes the data in the cache element occurs, a current value of the first counter is accumulated. When a write or other cache access that modifies data in the cache element occurs, the first counter is reset. At the end of an evaluation period, the value in a total counter approximates the number of clock cycles during which data that was consumed was vulnerable. Dividing this value by the number of clock cycles approximates the vulnerability of this cache element. The vulnerability for a subset of all cache elements may be measured and extrapolated to obtain an estimate for the vulnerability of the cache memory as a whole.

Vulnerability estimation for cache memory

Embodiments of apparatuses, methods, and storage medium associated with selectively providing error correction to memory are disclosed herein. In one instance, an apparatus may include a memory controller configured to control access to a non-volatile memory having storage locations. The controller may be configured to provide a first error correction arrangement to provide a first level of error correction capability for data stored in the non-volatile memory. The memory controller may include a control/error correction block configured to provide a second error correction arrangement with a second level of error correction capability for data stored in the non-volatile memory. The second level of error correction capability enables correction of at least one bit error more than the first level. The memory controller may be configured to selectively employ the second error correction arrangement to complement the first error correction arrangement. Other embodiments may be described and claimed.

Selective provision of error correction for memory

In one embodiment, a processor includes at least one functional block and banking logic. The banking logic may be to determine an average reliability metric associated with the at least one functional block. The banking logic may also be to, if the average reliability metric exceeds a required level, implement a reduced reliability mode in the at least one functional block, where the reduced reliability mode is associated with a reduction in the average reliability metric. Other embodiments are described and claimed.

Growing community of inventors

Cardedeu, Spain

Enric Herrero Abellanas