Abstract: Neural network processing hardware using parallel computational architectures with reconfigurable core-level and vector-level parallelism is provided. In various embodiments, a neural network model memory is adapted to store a neural network model comprising a plurality of layers. Each layer has at least one dimension and comprises a plurality of synaptic weights. A plurality of neural cores is provided. Each neural core includes a computation unit and an activation memory. The computation unit is adapted to apply a plurality of synaptic weights to a plurality of input activations to produce a plurality of output activations. The computation unit has a plurality of vector units. The activation memory is adapted to store the input activations and the output activations. The system is adapted to partition the plurality of cores into a plurality of partitions based on dimensions of the layer and the vector units.

Abstract: Learned step size quantization in artificial neural network is provided. In various embodiments, a system comprises an artificial neural network and a computing node. The artificial neural network comprises: a quantizer having a configurable step size, the quantizer adapted to receive a plurality of input values and quantize the plurality of input values according to the configurable step size to produce a plurality of quantized input values, at least one matrix multiplier configured to receive the plurality of quantized input values from the quantizer and to apply a plurality of weights to the quantized input values to determine a plurality of output values having a first precision, and a multiplier configured to scale the output values to a second precision.

Abstract: Instruction distribution in an array of neural network cores is provided. In various embodiments, a neural inference chip is initialized with core microcode. The chip comprises a plurality of neural cores. The core microcode is executable by the neural cores to execute a tensor operation of a neural network. The core microcode is distributed to the plurality of neural cores via an on-chip network. The core microcode is executed synchronously by the plurality of neural cores to compute a neural network layer.

Abstract: Long-short term memory (LSTM) cells on spiking neuromorphic hardware are provided. In various embodiments, such systems comprise a spiking neurosynaptic core. The neurosynaptic core comprises a memory cell, an input gate operatively coupled to the memory cell and adapted to selectively admit an input to the memory cell, and an output gate operatively coupled to the memory cell an adapted to selectively release an output from the memory cell. The memory cell is adapted to maintain a value in the absence of input.

Abstract: Hardware neural network processors, are provided. A neural core includes a weight memory, an activation memory, a vector-matrix multiplier, and a vector processor. The vector-matrix multiplier is adapted to receive a weight matrix from the weight memory, receive an activation vector from the activation memory, and compute a vector-matrix multiplication of the weight matrix and the activation vector. The vector processor is adapted to receive one or more input vector from one or more vector source and perform one or more vector functions on the one or more input vector to yield an output vector. In some embodiments a programmable controller is adapted to configure and operate the neural core.

Abstract: Core utilization optimization by dividing computational blocks across neurosynaptic cores is provided. In some embodiments, a neural network description describing a neural network is read. The neural network comprises a plurality of functional units on a plurality of cores. A functional unit is selected from the plurality of functional units. The functional unit is divided into a plurality of subunits. The plurality of subunits are connected to the neural network in place of the functional unit. The plurality of functional units and the plurality of subunits are reallocated between the plurality of cores. One or more unused cores are removed from the plurality of cores. An optimized neural network description is written based on the reallocation.

Abstract: An event-driven neural network including a plurality of interconnected core circuits is provided. Each core circuit includes an electronic synapse array that has multiple digital synapses interconnecting a plurality of digital electronic neurons. A synapse interconnects an axon of a pre-synaptic neuron with a dendrite of a post-synaptic neuron. A neuron integrates input spikes and generates a spike event in response to the integrated input spikes exceeding a threshold. Each core circuit also has a scheduler that receives a spike event and delivers the spike event to a selected axon in the synapse array based on a schedule for deterministic event delivery.

Abstract: Neural inference chips are provided. A neural core of the neural inference chip comprises a vector-matrix multiplier; a vector processor; and an activation unit operatively coupled to the vector processor. The vector-matrix multiplier, vector processor, and/or activation unit is adapted to operate at variable precision.

Abstract: Neural inference chips for computing neural activations are provided. In various embodiments, a neural inference chip comprises at least one neural core, a memory array, an instruction buffer, and an instruction memory. The instruction buffer has a position corresponding to each of a plurality of elements of the memory array. The instruction memory provides at least one instruction to the instruction buffer. The instruction buffer advances the at least one instruction between positions in the instruction buffer. The instruction buffer provides the at least one instruction to at least one of the plurality of elements of the memory array from its associated position in the instruction buffer when the memory of the at least one of the plurality of elements contains data associated with the at least one instruction. Each element of the memory array provides a data block from its memory to its horizontal buffer in response to the arrival of an associated instruction from the instruction buffer.

Abstract: Hardware neural network processors, are provided. A neural core includes a weight memory, an activation memory, a vector-matrix multiplier, and a vector processor. The vector-matrix multiplier is adapted to receive a weight matrix from the weight memory, receive an activation vector from the activation memory, and compute a vector-matrix multiplication of the weight matrix and the activation vector. The vector processor is adapted to receive one or more input vector from one or more vector source and perform one or more vector functions on the one or more input vector to yield an output vector. In some embodiments a programmable controller is adapted to configure and operate the neural core.

Abstract: The present invention relates to unsupervised, supervised and reinforced learning via spiking computation. The neural network comprises a plurality of neural modules. Each neural module comprises multiple digital neurons such that each neuron in a neural module has a corresponding neuron in another neural module. An interconnection network comprising a plurality of edges interconnects the plurality of neural modules. Each edge interconnects a first neural module to a second neural module, and each edge comprises a weighted synaptic connection between every neuron in the first neural module and a corresponding neuron in the second neural module.

Abstract: Embodiments of the invention provide a neural network comprising multiple functional neural core circuits, and a dynamically reconfigurable switch interconnect between the functional neural core circuits. The interconnect comprises multiple connectivity neural core circuits. Each functional neural core circuit comprises a first and a second core module. Each core module comprises a plurality of electronic neurons, a plurality of incoming electronic axons, and multiple electronic synapses interconnecting the incoming axons to the neurons. Each neuron has a corresponding outgoing electronic axon. In one embodiment, zero or more sets of connectivity neural core circuits interconnect outgoing axons in a functional neural core circuit to incoming axons in the same functional neural core circuit.

Abstract: A neural inference chip is provided, including at least one neural inference core. The at least one neural inference core is adapted to apply a plurality of synaptic weights to a plurality of input activations to produce a plurality of intermediate outputs. The at least one neural inference core comprises a plurality of activation units configured to receive the plurality of intermediate outputs and produce a plurality of activations. Each of the plurality of activation units is configured to apply a configurable activation function to its input. The configurable activation function has at least a re-ranging term and a scaling term, the re-ranging term determining the range of the activations and the scaling term determining the scale of the activations. Each of the plurality of activations units is configured to obtain the re-ranging term and the scaling term from one or more look up tables.

Abstract: Embodiments of the invention relate to a neural network system for simulating neurons of a neural model. One embodiment comprises a memory device that maintains neuronal states for multiple neurons, a lookup table that maintains state transition information for multiple neuronal states, and a controller unit that manages the memory device. The controller unit updates a neuronal state for each neuron based on incoming spike events targeting said neuron and state transition information corresponding to said neuronal state.

Abstract: Systems are provided that can produce symbolic and numeric representations of the neural network outputs, such that these outputs can be used to validate correctness of the implementation of the neural network. In various embodiments, a description of an artificial neural network containing no data-dependent branching is read. Based on the description of the artificial neural network, a symbolic representation is constructed of an output of the artificial neural network, the symbolic representation comprising at least one variable. The symbolic representation is compared to a ground truth symbolic representation, thereby validating the neural network system.

Abstract: Modular neural network computing apparatus are provided with distributed neural network storage. In various embodiments, a neural inference processor comprises a plurality of neural inference cores, at least one model network interconnecting the plurality of neural inference cores, and at least one activation network interconnecting the plurality of neural inference cores. Each of the plurality of neural inference cores comprises memory adapted to store input activations, output activations, and a neural network model. The neural network model comprises synaptic weights, neuron parameters, and neural network instructions. The at least one model network is configured to distribute the neural network model among the plurality of neural inference cores. Each of the plurality of neural inference cores is configured to apply the synaptic weights to input activations from its memory to produce a plurality of output activations to its memory.

Abstract: Simulation and validation of neural network systems is provided. In various embodiments, a description of an artificial neural network is read. A directed graph is constructed comprising a plurality of edges and a plurality of nodes, each of the plurality of edges corresponding to a queue and each of the plurality of nodes corresponding to a computing function of the neural network system. A graph state is updated over a plurality of time steps according to the description of the neural network, the graph state being defined by the contents of each of the plurality of queues. Each of a plurality of assertions is tested at each of the plurality of time steps, each of the plurality of assertions being a function of a subset of the graph state. Invalidity of the neural network system is indicated for each violation of one of the plurality of assertions.

Abstract: Neural network accelerator output ranking is provided. In various embodiments, a system comprises a data memory; a memory controller configured to access the data memory; a plurality of comparators configured in a tree; a register; and a two-way comparator. The memory controller is configured to provide a first plurality of values from the data memory to the comparator tree. The comparator tree is configured to perform a plurality of concurrent pairwise comparisons of the first plurality of values to arrive at a first greatest value of the first plurality of values. The two-way comparator is configured to output the greater of the greatest value from the comparator tree and a stored value from the register. The register is configured to store the output of the two-way comparator.

Abstract: Chips supporting constant time program control of nested loops are provided. In various embodiments, a chip comprises at least one arithmetic-logic computing unit and a controller operatively coupled to the at least one arithmetic-logic computing unit. The controller is configured according to a program configuration, the program configuration comprising at least one inner loop and at least one outer loop. The controller is configured to cause the at least one arithmetic computing unit to execute a plurality of operations according to the program configuration. The controller is configured to maintain at least a first loop counter and a second loop counter, the first loop counter configured to count a number of executed iterations of the at least one outer loop, and the second loop counter configured to count a number of executed iterations of the at least one inner loop.

Abstract: Conflict-free, stall-free, broadcast networks on neural inference chips are provided. In various embodiments, a neural inference chip comprises a plurality of network nodes and a network on chip interconnecting the plurality of network nodes. The network comprises at least one pair of directional paths. The paths of each pair have opposite directions and a common end. The network is configured to accept data at any of the plurality of nodes. The network is configured to propagate data along a first of the pair of directional paths from a source node to the common end of the pair of directional paths and along a second of the pair of directional paths from the common end of the pair of directional paths to one or more destination node.