Systems, apparatuses, and methods related to acceleration circuitry for posit operations are described. Signaling indicative of performance of an operation to write a first bit string to a first buffer resident on acceleration circuitry and a second bit string resident on the acceleration circuitry can be received at an DMA controller couplable to the acceleration circuitry. The acceleration circuitry can be configured to perform arithmetic operations, logical operations, or both on bit strings formatted in a unum or posit format. Signaling indicative of an arithmetic operation, a logical operation, or both, to be performed using the first and second bit strings can be transmitted to the acceleration circuitry. The arithmetic operation, the logical operation, or both can be performed via the acceleration circuitry and according to the signaling. Signaling indicative of a result of the arithmetic operation, the logical operation, or both can be transmitting to the DMA controller.

A processor module including a processor configured to share data with at least one further processor module processor; and a memory mapped peripheral configured to communicate with at least one further processor memory mapped peripheral to control the sharing of the data, wherein the memory mapped peripheral includes a sender part including a data request generator configured to output a data request indicator to the further processor module dependent on a data request register write signal from the processor; and an acknowledgement waiting signal generator configured to output an acknowledgement waiting signal to the processor dependent on a data acknowledgement signal from the further processor module, wherein the data request generator data request indicator is further dependent on the data acknowledgement signal and the acknowledgement waiting signal generator acknowledgement waiting signal is further dependent on the acknowledgement waiting register write signal.

A data processing system can include a first IC including one or more A/D converters that receive analog inputs from one or more sensors and generate corresponding digital data, a second IC including one or more processing elements that operate on the digital data, and communication circuitry, coupled between the one or more A/D converters and processing elements, that includes a packetizer on the first IC that receives samples and sample data from the one or more A/D converters and assembles each sample and corresponding sample data into a packet, a primary physical interface on the first IC that communicates the packet to a secondary physical interface on the second IC, and a de-packetizer that on the second IC that receives the packet, de-packetizes it, and delivers the sample and sample data to the one or more processing elements.

A data processing system can include a first IC including one or more A/D converters that receive analog inputs from one or more sensors and generate corresponding digital data, a second IC including one or more processing elements that operate on the digital data, and communication circuitry, coupled between the one or more A/D converters and processing elements, that includes a packetizer on the first IC that receives samples and sample data from the one or more A/D converters and assembles each sample and corresponding sample data into a packet, a primary physical interface on the first IC that communicates the packet to a secondary physical interface on the second IC, and a de-packetizer that on the second IC that receives the packet, de-packetizes it, and delivers the sample and sample data to the one or more processing elements.

A system that calibrates timing relationships between signals involved in performing write operations is described. This system includes a memory controller which is coupled to a set of memory chips, wherein each memory chip includes a phase detector configured to calibrate a phase relationship between a data-strobe signal and a clock signal received at the memory chip from the memory controller during a write operation. Furthermore, the memory controller is configured to perform one or more write-read-validate operations to calibrate a clock-cycle relationship between the data-strobe signal and the clock signal, wherein the write-read-validate operations involve varying a delay on the data-strobe signal relative to the clock signal by a multiple of a clock period.

A system that calibrates timing relationships between signals involved in performing write operations is described. This system includes a memory controller which is coupled to a set of memory chips, wherein each memory chip includes a phase detector configured to calibrate a phase relationship between a data-strobe signal and a clock signal received at the memory chip from the memory controller during a write operation. Furthermore, the memory controller is configured to perform one or more write-read-validate operations to calibrate a clock-cycle relationship between the data-strobe signal and the clock signal, wherein the write-read-validate operations involve varying a delay on the data-strobe signal relative to the clock signal by a multiple of a clock period.

A method and apparatus for synchronizing a timebase is disclosed. A timebase management circuit includes limit circuitry, in a first clock domain, which generates, based on a global timebase, an initial timebase limit. The timebase management circuit includes, in a second clock domain, adjustment circuitry that generates an adjusted timebase limit based on the initial timebase limit. A storage circuit in the second clock domain stores a local timebase. Update circuitry, coupled to an output of the storage circuit, generates an updated local timebase using a clock signal in the second clock domain, wherein the updated local timebase is subject to the adjusted timebase limit.

A method of exploiting activation sparsity in deep neural networks is described. The method includes retrieving an activation tensor and a weight tensor where the activation tensor is a sparse activation tensor. The method also includes generating a compressed activation tensor comprising non-zero activations of the activation tensor, where the compressed activation tensor has fewer columns than the activation tensor. The method further includes processing the compressed activation tensor and the weight tensor to generate an output tensor.

A method of exploiting activation sparsity in deep neural networks is described. The method includes retrieving an activation tensor and a weight tensor where the activation tensor is a sparse activation tensor. The method also includes generating a compressed activation tensor comprising non-zero activations of the activation tensor, where the compressed activation tensor has fewer columns than the activation tensor. The method further includes processing the compressed activation tensor and the weight tensor to generate an output tensor.

A digital communication interface includes a deserializer module, a gearbox module, and a parallel communication channel connecting the gearbox module to the deserializer module. The deserializer module has a fixed deserialization factor. The gearbox module has a temporal translation factor to change bit-length of words received through the parallel communication channel to bit-length suitable for a downstream data path.