Abstract: A neural processor. In some embodiments, the processor includes a first tile, a second tile, a memory, and a bus. The bus may be connected to the memory, the first tile, and the second tile. The first tile may include: a first weight register, a second weight register, an activations buffer, a first multiplier, and a second multiplier. The activations buffer may be configured to include: a first queue connected to the first multiplier and a second queue connected to the second multiplier. The first queue may include a first register and a second register adjacent to the first register, the first register being an output register of the first queue. The first tile may be configured: in a first state: to multiply, in the first multiplier, a first weight by an activation from the output register of the first queue, and in a second state: to multiply, in the first multiplier, the first weight by an activation from the second register of the first queue.

Abstract: An image sensor includes a time-resolving sensor and a processor. The time-resolving sensor outputs a first signal and a second signal pair in response detecting one or more photons that have been reflected from an object. A first ratio of a magnitude of the first signal to a sum of the magnitude of the first signal and a magnitude of the second signal is proportional to a time of flight of the one or more detected photons. A second ratio of the magnitude of the second signal to the sum of the magnitude of the first signal and the magnitude of the second signal is proportional to the time of flight of the one or more detected photons. The processor determines a surface reflectance of the object where the light pulse has been reflected based on the first signal and the second signal pair and may generate a grayscale image.

Abstract: Two-dimensional (2D) color information and 3D-depth information are concurrently obtained from a 2D pixel array. The 2D pixel array is arranged in a first group of a plurality of rows. A second group of rows of the array are operable to generate 2D-color information and pixels of a third group of the array are operable to generate 3D-depth information. The first group of rows comprises a first number of rows, the second group of rows comprises a second number of rows that is equal to or less than the first number of rows, and the third group of rows comprises a third number of rows that is equal to or less than the second number of rows. In an alternating manner, 2D-color information is received from a row selected from the second group of rows and 3D-depth information is received from a row selected from the third group of rows.

Abstract: A processor to perform inference on deep learning neural network models. In some embodiments, the process includes: a first tile, a second tile, a memory, and a bus, the bus being connected to: the memory, the first tile, and the second tile, the first tile including: a first weight register, a second weight register, an activations cache, a shuffler, an activations buffer, a first multiplier, and a second multiplier, the activations buffer being configured to include: a first queue connected to the first multiplier, and a second queue connected to the second multiplier, the activations cache including a plurality of independent lanes, each of the independent lanes being randomly accessible, the first tile being configured: to receive a tensor including a plurality of two-dimensional arrays, each representing one color component of the image; and to perform a convolution of a kernel with one of the two-dimensional arrays.

Abstract: A client device configured with a neural network includes a processor, a memory, a user interface, a communications interface, a power supply and an input device, wherein the memory includes a trained neural network received from a server system that has trained and configured the neural network for the client device. A server system and a method of training a neural network are disclosed.

Abstract: A neural processor. In some embodiments, the processor includes a first tile, a second tile, a memory, and a bus. The bus may be connected to the memory, the first tile, and the second tile. The first tile may include: a first weight register, a second weight register, an activations buffer, a first multiplier, and a second multiplier. The activations buffer may be configured to include: a first queue connected to the first multiplier and a second queue connected to the second multiplier. The first queue may include a first register and a second register adjacent to the first register, the first register being an output register of the first queue. The first tile may be configured: in a first state: to multiply, in the first multiplier, a first weight by an activation from the output register of the first queue, and in a second state: to multiply, in the first multiplier, the first weight by an activation from the second register of the first queue.

Abstract: A neural processor. In some embodiments, the processor includes a first tile, a second tile, a memory, and a bus. The bus may be connected to the memory, the first tile, and the second tile. The first tile may include: a first weight register, a second weight register, an activations buffer, a first multiplier, and a second multiplier. The activations buffer may be configured to include: a first queue connected to the first multiplier and a second queue connected to the second multiplier. The first queue may include a first register and a second register adjacent to the first register, the first register being an output register of the first queue. The first tile may be configured: in a first state: to multiply, in the first multiplier, a first weight by an activation from the output register of the first queue, and in a second state: to multiply, in the first multiplier, the first weight by an activation from the second register of the first queue.

Abstract: A neural processor. In some embodiments, the processor includes a first tile, a second tile, a memory, and a bus. The bus may be connected to the memory, the first tile, and the second tile. The first tile may include: a first weight register, a second weight register, an activations buffer, a first multiplier, and a second multiplier. The activations buffer may be configured to include: a first queue connected to the first multiplier and a second queue connected to the second multiplier. The first queue may include a first register and a second register adjacent to the first register, the first register being an output register of the first queue. The first tile may be configured: in a first state: to multiply, in the first multiplier, a first weight by an activation from the output register of the first queue, and in a second state: to multiply, in the first multiplier, the first weight by an activation from the second register of the first queue.

Abstract: An N×N multiplier may include a N/2×N first multiplier, a N/2×N/2 second multiplier, and a N/2×N/2 third multiplier. The N×N multiplier receives two operands to multiply. The first, second and/or third multipliers are selectively disabled if an operand equals zero or has a small value. If the operands are both less than 2N/2, the second or the third multiplier are used to multiply the operands. If one operand is less than 2N/2 and the other operand is equal to or greater than 2N/2, the first multiplier is used or the second and third multipliers are used to multiply the operands. If both operands are equal to or greater than 2N/2, the first, second and third multipliers are used to multiply the operands.

Abstract: A neural processor. In some embodiments, the processor includes a first tile, a second tile, a memory, and a bus. The bus may be connected to the memory, the first tile, and the second tile. The first tile may include: a first weight register, a second weight register, an activations buffer, a first multiplier, and a second multiplier. The activations buffer may be configured to include: a first queue connected to the first multiplier and a second queue connected to the second multiplier. The first queue may include a first register and a second register adjacent to the first register, the first register being an output register of the first queue. The first tile may be configured: in a first state: to multiply, in the first multiplier, a first weight by an activation from the output register of the first queue, and in a second state: to multiply, in the first multiplier, the first weight by an activation from the second register of the first queue.

Abstract: A neural processor. In some embodiments, the processor includes a first tile, a second tile, a memory, and a bus. The bus may be connected to the memory, the first tile, and the second tile. The first tile may include: a first weight register, a second weight register, an activations buffer, a first multiplier, and a second multiplier. The activations buffer may be configured to include: a first queue connected to the first multiplier and a second queue connected to the second multiplier. The first queue may include a first register and a second register adjacent to the first register, the first register being an output register of the first queue. The first tile may be configured: in a first state: to multiply, in the first multiplier, a first weight by an activation from the output register of the first queue, and in a second state: to multiply, in the first multiplier, the first weight by an activation from the second register of the first queue.

Abstract: Using the same image sensor to capture a two-dimensional (2D) image and three-dimensional (3D) depth measurements for a 3D object. A laser point-scans the surface of the object with light spots, which are detected by a pixel array in the image sensor to generate the 3D depth profile of the object using triangulation. Each row of pixels in the pixel array forms an epipolar line of the corresponding laser scan line. Timestamping provides a correspondence between the pixel location of a captured light spot and the respective scan angle of the laser to remove any ambiguity in triangulation. An Analog-to-Digital Converter (ADC) in the image sensor operates as a Time-to-Digital (TDC) converter to generate timestamps. A timestamp calibration circuit is provided on-board to record the propagation delay of each column of pixels in the pixel array and to provide necessary corrections to the timestamp values generated during 3D depth measurements.

Abstract: A system for calculating. A scratch memory is connected to a plurality of configurable processing elements by a communication fabric including a plurality of configurable nodes. The scratch memory sends out a plurality of streams of data words. Each data word is either a configuration word used to set the configuration of a node or of a processing element, or a data word carrying an operand or a result of a calculation. Each processing element performs operations according to its current configuration and returns the results to the communication fabric, which conveys them back to the scratch memory.

Abstract: A data compressor a zero-value remover, a zero bit mask generator, a non-zero values packer, and a row-pointer generator. The zero-value remover receives 2N bit streams of values and outputs 2N non-zero-value bit streams having zero values removed from each respective bit stream. The zero bit mask generator receives the 2N bit streams of values and generates a zero bit mask for a predetermined number of values of each bit stream in which each zero bit mask indicates a location of a zero value in the predetermined number of values corresponding to the zero bit mask. The non-zero values packer receives the 2N non-zero-value bit streams and forms a group of packed non-zero values. The row-pointer generator that generates a row-pointer for each group of packed non-zero values.

Abstract: A system and a method of quantizing a pre-trained neural network, includes determining by a layer/channel bit-width determiner for each layer or channel of the pre-trained neural network a minimum quantization noise for the layer or the channel for each master bit-width value in a predetermined set of master bit-width values; and selecting by a bit-width selector for the layer or the channel the master bit-width value having the minimum quantization noise for the layer or the channel. In one embodiment, the minimum quantization noise for the layer or the channel is based on a square of a range of weights for the layer or the channel that is multiplied by a constant to a negative power of a current master bit-width value.

Abstract: A multichannel data packer includes a plurality of two-input multiplexers and a controller. The plurality of two-input multiplexers is arranged in 2N rows and N columns in which N is an integer greater than 1. Each input of a multiplexer in a first column receives a respective bit stream of 2N channels of bit streams. Each respective bit stream includes a bit-stream length based on data in the bit stream. The multiplexers in a last column output 2N channels of packed bit streams each having a same bit-stream length. The controller controls the plurality of multiplexers so that the multiplexers in the last column output the 2N channels of bit streams that each has the same bit-stream length.

Abstract: A data-sparsity homogenizer includes a plurality of multiplexers and a controller. The plurality of multiplexers receives 2N bit streams of non-homogenous sparse data in which the non-homogenous sparse data includes non-zero value data clumped together. The plurality of multiplexers is arranged in 2N rows and N columns. Each input of a multiplexer in a first column receives a respective bit stream of the 2N bit streams of non-homogenized sparse data, and the multiplexers in a last column output 2N bit streams of sparse data that is more homogenous than the non-homogenous sparse data of the 2N bit streams. The controller controls the plurality of multiplexers so that the multiplexers in the last column output the 2N channels of bit streams of sparse data that is more homogeneous than the non-homogenous sparse data of the 2N bit streams.

Abstract: An N×N multiplier may include a N/2×N first multiplier, a N/2×N/2 second multiplier, and a N/2×N/2 third multiplier. The N×N multiplier receives two operands to multiply. The first, second and/or third multipliers are selectively disabled if an operand equals zero or has a small value. If the operands are both less than 2N/2, the second or the third multiplier are used to multiply the operands. If one operand is less than 2N/2 and the other operand is equal to or greater than 2N/2, the first multiplier is used or the second and third multipliers are used to multiply the operands. If both operands are equal to or greater than 2N/2, the first, second and third multipliers are used to multiply the operands.

Abstract: A system for calculating. A scratch memory is connected to a plurality of configurable processing elements by a communication fabric including a plurality of configurable nodes. The scratch memory sends out a plurality of streams of data words. Each data word is either a configuration word used to set the configuration of a node or of a processing element, or a data word carrying an operand or a result of a calculation. Each processing element performs operations according to its current configuration and returns the results to the communication fabric, which conveys them back to the scratch memory.

Abstract: A client device configured with a neural network includes a processor, a memory, a user interface, a communications interface, a power supply and an input device, wherein the memory includes a trained neural network received from a server system that has trained and configured the neural network for the client device. A server system and a method of training a neural network are disclosed.