Disclosed is an Avalon-to-Axi4 bus conversion method, including: in case that an Avalon bus is an Avalon_st bus, receiving Avalon_st bus data, performing a logical process on the received Avalon_st bus data, and then outputting corresponding Axi4_st bus data; and in case that the Avalon bus is an Avalon_mm bus, receiving a signal transmitted by each channel of the Avalon_mm bus, framing and storing the signal in asynchronous First Input First Output (FIFO), and in case that a device corresponding to an Axi4 bus is ready, reading the signal from the asynchronous FIFO, and outputting the signal to a corresponding channel of the Axi4 bus according to a timing relationship of the Axi4 bus.

This disclosure relates to methods, non-transitory computer readable media, and systems that asynchronously train a machine learning model across client devices that implement local versions of the model while preserving client data privacy. To train the model across devices, in some embodiments, the disclosed systems send global parameters for a global machine learning model from a server device to client devices. A subset of the client devices uses local machine learning models corresponding to the global model and client training data to modify the global parameters. Based on those modifications, the subset of client devices sends modified parameter indicators to the server device for the server device to use in adjusting the global parameters. By utilizing the modified parameter indicators (and not client training data), in certain implementations, the disclosed systems accurately train a machine learning model without exposing training data from the client device.

A method for providing an asynchronous execution queue for accelerator hardware includes replacing a malloc operation in an execution queue to be sent to an accelerator with an asynchronous malloc operation that returns a unique reference pointer. Execution of the asynchronous malloc operation in the execution queue by the accelerator allocates a requested memory size and adds an entry to a look-up table accessible by the accelerator that maps the reference pointer to a corresponding memory address.

An information processing apparatus includes a processor, a memory, a memory controller, and a storage. The memory serves as a main memory of the processor. The memory controller controls a first access from the processor to the memory, a second access to the memory that is performed without being synchronized with the first access, and processing related to memory dump acquisition. The storage stores, upon performing the second access, a memory dump of data stored in the memory, according to an instruction given by the memory controller.

A method for stack timing adjustment for serial communications is provided. The method includes receiving a USB communication, decoding the USB communication into UART frames, and adjusting the timing of the UART frames according to a serial protocol.

In a data processing device including two sets of circuit pairs which are respectively duplicated in two clock domains which are asynchronous to each other, an asynchronous transfer circuit that transfers a payload signal is provided between the two sets of circuit pairs. The asynchronous transfer circuit includes two sets of a pair of bridge circuits which are respectively connected to the two sets of circuit pairs, and asynchronously transfers the payload signal and a control signal indicating a timing at which the payload signal is stable on a reception side. The two sets of a pair of bridge circuits and the payload signals can be duplicated, but the control signal is not duplicated, and the received payload signal is used for timing control to supply an expected same time difference, to the pair of duplicated circuits. This enables asynchronous transfer between circuits duplicated in the asynchronous clock domains.

Aspects of the disclosure provide an interface between a host and a multi-plane flash memory. For example, the interface can include a first storage unit, a second storage unit and a controller. The first storage unit can be configured to receive and store a first plane pipeline command issued from the host, and output the first plane pipeline command to a first plane of the flash memory. The second storage unit can be configured to receive and store a second plane pipeline command issued from the host, and output the second plane pipeline command to a second plane of the flash memory. The controller can be electrically connected to the first storage unit and the second storage unit, and configured to output the first and second plane pipeline commands to the first and second planes, respectively, when no read process is performed on the first plane and the second plane.

A semiconductor memory device includes a substrate that has a first main surface and a second main surface opposite to the first main surface, a first semiconductor chip which is mounted on the first main surface and includes a first register, a plurality of first input/output (IO) terminals, and a first circuit connected between the first IO terminals and the first register, and a second semiconductor chip which is mounted on the second main surface and includes a second register, a plurality of second input/output (IO) terminals, and a second circuit connected between the second IO terminals and the second register. The second circuit is connected to the second IO terminals through input lines and to the second register through output lines, and is configured to change a connection path between the input lines and the output lines in response to a connection change command.

Methods, apparatus, systems, and articles of manufacture are disclosed to steal work in heterogeneous computing systems. An apparatus includes load balancing circuitry to obtain tasks from a workload by encoding minimum and maximum index ranges of a data parallel operation, allocate a first task from the workload to a first work queue based on a first capability of first computation circuitry, the first computation circuitry to process the first task in the first work queue, and allocate a second task from the workload to a second work queue, second computation circuitry to process the second task in the second work queue. The apparatus further includes first work stealer logic to steal the second task from the second work queue using an atomic operation to access the second work queue.

Disclosed is an Avalon-to-Axi4 bus conversion method, including: in case that an Avalon bus is an Avalon_st bus, receiving Avalon_st bus data, performing a logical process on the received Avalon_st bus data, and then outputting corresponding Axi4_st bus data; and in case that the Avalon bus is an Avalon_mm bus, receiving a signal transmitted by each channel of the Avalon_mm bus, framing and storing the signal in asynchronous First Input First Output (FIFO), and in case that a device corresponding to an Axi4 bus is ready, reading the signal from the asynchronous FIFO, and outputting the signal to a corresponding channel of the Axi4 bus according to a timing relationship of the Axi4 bus.