Abstract: A distributed deep neural net (DNN) utilizing a distributed, tile-based architecture implemented on a semiconductor package. The package includes multiple chips, each with a central processing element, a global memory buffer, and processing elements. Each processing element includes a weight buffer, an activation buffer, and multiply-accumulate units to combine, in parallel, the weight values and the activation values.

Abstract: IR drop predictions are obtained using a maximum convolutional neural network. A circuit structure is partitioned into a grid. For cells of the circuit structure in sub-intervals of a clock period, power consumption of the cell is amortized into a set of grid tiles that include portions of the cell, thus forming a set of power maps. The power maps are applied to a neural network to generate IR drop predictions for the circuit structure.

Abstract: One embodiment of a computer-implemented method for processing a neural network comprises receiving a first quantized matrix that corresponds to a portion of a multi-dimensional input tensor and has been quantized based on a first scale factor; and performing one or more computational operations using the first quantized matrix and the first scale factor to generate one or more data values that correspond to a first portion of a multi-dimensional output tensor.

Abstract: A distributed deep neural net (DNN) utilizing a distributed, tile-based architecture includes multiple chips, each with a central processing element, a global memory buffer, and a plurality of additional processing elements. Each additional processing element includes a weight buffer, an activation buffer, and vector multiply-accumulate units to combine, in parallel, the weight values and the activation values using stationary data flows.

Abstract: A distributed deep neural net (DNN) utilizing a distributed, tile-based architecture includes multiple chips, each with a central processing element, a global memory buffer, and a plurality of additional processing elements. Each additional processing element includes a weight buffer, an activation buffer, and vector multiply-accumulate units to combine, in parallel, the weight values and the activation values using stationary data flows.

Abstract: Solutions improving efficiency of Softmax computation applied for efficient deep learning inference in transformers and other neural networks. The solutions utilize a reduced-precision implementation of various operations in Softmax, replacing ex with 2x to reduce instruction overhead associated with computing ex, and replacing floating point max computation with integer max computation. Further described is a scalable implementation that decomposes Softmax into UnNormalized Softmax and Normalization operations.

Abstract: A graph neural network for average power estimation of netlists is trained with register toggle rates over a power window from an RTL simulation and gate level netlists as input features. Combinational gate toggle rates are applied as labels. The trained graph neural network is then applied to infer combinational gate toggle rates over a different power window of interest and/or different netlist.

Abstract: IR drop predictions are obtained using a maximum convolutional neural network. A circuit structure is partitioned into a grid. For cells of the circuit structure in sub-intervals of a clock period, power consumption of the cell is amortized into a set of grid tiles that include portions of the cell, thus forming a set of power maps. The power maps are applied to a neural network to generate IR drop predictions for the circuit structure.

Abstract: A distributed deep neural net (DNN) utilizing a distributed, tile-based architecture includes multiple chips, each with a central processing element, a global memory buffer, and a plurality of additional processing elements. Each additional processing element includes a weight buffer, an activation buffer, and vector multiply-accumulate units to combine, in parallel, the weight values and the activation values using stationary data flows.

Abstract: Techniques to improve the accuracy and speed for detection and remediation of difficult to test nodes in a circuit design netlist. The techniques utilize improved netlist representations, test point insertion, and trained neural networks.

Abstract: Techniques to improve the accuracy and speed for detection and remediation of difficult to test nodes in a circuit design netlist. The techniques utilize improved netlist representations, test point insertion, and trained neural networks.

Abstract: A distributed deep neural net (DNN) utilizing a distributed, tile-based architecture implemented on a semiconductor package. The package includes multiple chips, each with a central processing element, a global memory buffer, and processing elements. Each processing element includes a weight buffer, an activation buffer, and multiply-accumulate units to combine, in parallel, the weight values and the activation values.

Abstract: A neural network including an embedding layer to receive a gate function vector and an embedding width and alter a shape of the gate function vector by the embedding width, a concatenator to receive a gate feature input vector and concatenate the gate feature input vector with the gate function vector altered by the embedding width, a convolution layer to receive a window size, stride, and output feature size and generate an output convolution vector with a shape based on a length of the gate function vector, the window size of the convolution layer, and the output feature size of the convolution layer, and a fully connected layer to reduce the gate output convolution vector to a final path delay output.

Abstract: A neural network including an embedding layer to receive a gate function vector and an embedding width and alter a shape of the gate function vector by the embedding width, a concatenator to receive a gate feature input vector and concatenate the gate feature input vector with the gate function vector altered by the embedding width, a convolution layer to receive a window size, stride, and output feature size and generate an output convolution vector with a shape based on a length of the gate function vector, the window size of the convolution layer, and the output feature size of the convolution layer, and a fully connected layer to reduce the gate output convolution vector to a final path delay output.

Abstract: Tracing command execution in a data processing system having a host processor and a co-processor. The host processor maintains a record of a plurality of commands for the co-processor, storing each of the plurality of commands is stored in a command queue. Hardware trace logic is provided to store one or more events based, at least in part, on transfer of the plurality of commands to a small memory. Software is executed to store the one or more events to a main memory, wherein the one or more events are aggregated into a single memory trace within the main memory.

Abstract: Disclosed are methods and systems for dynamically determining data-transfer paths. The data-transfer paths are dynamically determined in response to an instruction that facilitates data transfer among execution lanes in an integrated-circuit processing device operable to execute operations in parallel. In addition, embodiments include an integrated-circuit processing device operable to execute operations in parallel, including the capability of providing confirmation information to potential source lanes, the confirmation information indicating whether the potential source lanes may send data to requested destination lanes during a data-transfer interval.

Abstract: A method of operation within an integrated-circuit processing device having an enhanced combined-arithmetic capability. In response to an instruction indicating a combined arithmetic operation, the processor generates a dot-product of first and second operands, adds the dot-product to an accumulated value, and then outputs the sum of the accumulated value and the dot-product.

Abstract: Disclosed are methods and systems for dynamically determining data-transfer paths. The data-transfer pats are determined in response to an instruction that facilitates data transfer among execution lanes in an integrated-circuit processing device operable to execute operations in parallel.

Abstract: A method of operation within an integrated-circuit processing device having a plurality of execution lanes. Upon receiving an instruction to exchange data between the execution lanes, respective requests from the execution lanes are examined to determine a set of the execution lanes that may send data to one or more others of the execution lanes during a first interval. Each execution lane within the set of the execution lanes is signaled to indicate that the execution lane may send data to the one or others of the execution lanes.

Abstract: An application programming interface is disclosed for loading and storing multidimensional arrays of data between a data parallel processing unit and an external memory. Physical addresses reference the external memory and define two-dimensional arrays of data storage locations corresponding to data records. The data parallel processing unit has multiple processing lanes to parallel process data records residing in respective register files. The interface comprises an X-dimension function call parameter to define an X-dimension in the memory array corresponding to a record for one lane and a Y-dimension function call parameter to define a Y-dimension in the memory array corresponding to the record for one lane. The X-dimension and Y-dimension function call parameters cooperate to generate memory accesses corresponding to the records.