Abstract: Systems, methods and apparatuses of distributed computing based on memory as a service are described. For example, a set of networked computing devices can each be configured to execute an application that accesses memory using a virtual memory address region. Each respective device can map the virtual memory address region to the local memory for a first period of time during which the application is being executed in the respective device, map the virtual memory address region to a local memory of a remote device in the group for a second period of time after starting the application in the respective device and before terminating the application in the respective device, and request the remote device to process data in the virtual memory address region during at least the second period of time.

Abstract: Methods, systems, and devices for a magnetic cache for a memory device are described. Magnetic storage elements (e.g., magnetic memory cells, such as spin-transfer torque (STT) memory cells or magnetic tunnel junction (MTJ) memory cells) may be configured to act as a cache for a memory array, where the memory array includes a different type of memory cells. The magnetic storage elements may be inductively coupled to access lines for the memory array. Based on this inductive coupling, when a memory value is written to or read from a memory cell of the array, the memory value may concurrently be written to a magnetic storage element based on associated current through an access line used to write or read the memory cell. Subsequent read requests may be executed by reading the memory value from the magnetic storage element rather than from the memory cell of the array.

Abstract: An example system implementing a processing-in-memory pipeline includes: a memory array to store a plurality of look-up tables (LUTs) and data; a control block coupled to the memory array, the control block to control a computational pipeline by activating one or more LUTs of the plurality of LUTs; and a logic array coupled to the memory array and the control block, the logic array to perform, based on control inputs received from the control block, logic operations on the activated LUTs and the data.

Abstract: Systems and methods for implementing shadow computations in base stations. The systems and methods can include a method including initiating, at a base station (such as a cellular base station), a shadow computation of a main computation executing for a mobile device. The main computation can include a computational task, and the shadow computation can be at least a part of or a derivative of the main computation. The method can also include executing, by the base station, the shadow computation.

Abstract: Methods, apparatuses, and systems for in- or near-memory processing are described. Strings of bits (e.g., vectors) may be fetched and processed in logic of a memory device without involving a separate processing unit. Operations (e.g., arithmetic operations) may be performed on numbers stored in a bit-parallel way during a single sequence of clock cycles. Arithmetic may thus be performed in a single pass as numbers are bits of two or more strings of bits are fetched and without intermediate storage of the numbers. Vectors may be fetched (e.g., identified, transmitted, received) from one or more bit lines. Registers of a memory array may be used to write (e.g., store or temporarily store) results or ancillary bits (e.g., carry bits or carry flags) that facilitate arithmetic operations. Circuitry near, adjacent, or under the memory array may employ XOR or AND (or other) logic to fetch, organize, or operate on the data.

Abstract: The present disclosure is directed to systems and methods for a memory device such as, for example, a Processing-In-Memory Device that is configured to perform multiplication operations in memory using a popcount operation. A multiplication operation may include a summation of multipliers being multiplied with corresponding multiplicands. The inputs may be arranged in particular configurations within a memory array. Sense amplifiers may be used to perform the popcount by counting active bits along bit lines. One or more registers may accumulate results for performing the multiplication operations.

Abstract: Methods, systems, and devices related to write broadcast operations associated with a memory device are described. In one example, a memory device in accordance with the described techniques may include a memory array, a sense amplifier array, and a signal development cache configured to store signals (e.g., cache signals, signal states) associated with logic states (e.g., memory states) that may be stored at the memory array (e.g., according to various read or write operations). The memory device may enable read broadcast operations. A read broadcast may occur from the memory array to multiple locations of the signal development cache, for example via one or more multiplexers.

Abstract: The disclosed embodiments include a memory array configured to store a membrane potential and a synaptic connection identifier of each of a plurality of neurons, a plurality of processors coupled to the memory array, the plurality of processors configured to: immediately perform a search and match operation in the memory array upon receiving a spike message identifying relevant synaptic connections in the memory array, generate a bitmask signifying a first source neuron identifier having a match to a second source neuron identifier in the memory array, perform a synaptic integration and a long-time depression computation on a subset of spike messages including the first spike message, update membrane potentials of the plurality of neurons upon receiving an indication that all the spike messages identified in a barrier message have been received in the memory array, generate a new spike message, and transmit the new spike message to a network.

Abstract: The disclosed embodiments relate to per-process configuration caches in storage devices.

Abstract: Systems, methods, and apparatus related to memory devices. In one approach, a memory device includes memories that store data received from a host device. The memory device includes a memory interface to the host device. The memory device further includes one or more processing devices to perform, using a portion of the data stored in one or more of the memories, computations for a neural network. An output of the neural network is stored in one of the memories. The memory device has a controller that controls memory access by the host device and the processing devices to avoid a conflict. The memory device communicates with the host device over the memory interface using a DRAM bus protocol. This communication includes sending the output of the neural network to the host device.

Abstract: Methods, apparatuses, and systems for in- or near-memory processing are described. Spiking events in a spiking neural network may be processed via a memory system. A memory system may store a group of destination neurons, and at each time interval in a series of time intervals of a spiking neural network (SNN), pass through a group of pre-synaptic spike events from respective source neurons, wherein the group of pre-synaptic spike events are subsequently stored in memory.

Abstract: An example system implementing a processing-in-memory pipeline includes: a memory array to store data in a plurality of memory cells electrically coupled to a plurality of wordlines and a plurality of bitlines; a logic array coupled to the memory array, the logic array to implement configurable logic controlling the plurality of memory cells; and a control block coupled to the memory array and the logic array, the control block to control a computational pipeline to perform computations on the data by activating at least one of: one or more bitlines of the plurality of bitlines or one or more wordlines of the plurality of wordlines.

Abstract: Systems and methods for performing a pattern matching operation in a memory device are disclosed. The memory device may include a controller and memory arrays where the memory arrays store different patterns along bit lines. An input pattern is applied to the memory array(s) to determine whether the pattern is stored in the memory device. Word lines may be activated in series or in parallel to search for patterns within the memory array. The memory array may include memory cells that store binary digits, discrete values or analog values.

Abstract: A three-dimensional integrated circuit includes a first die structure having a first geometry. The first die structure includes a first region that operates with a first power density and a second region that operates with a second power density. The first power density is less than the second power density. The three-dimensional integrated circuit includes a second die structure having a second geometry. A stacked portion of the second die structure is aligned with the first region. The three-dimensional integrated circuit includes an additional die structure stacked with the first die structure and the second die structure. The additional die structure has the first geometry or the second geometry.

Abstract: An artificial neural network that includes first subnetworks to implement known functions and second subnetworks to implement unknown functions is trained. The first subnetworks are trained separately and in parallel on corresponding known training datasets to determine first parameter values that define the first subnetworks. The first subnetworks are executing on a plurality of processing elements in a processing system. Input values from a network training data set are provided to the artificial neural network including the trained first subnetworks. Error values are generated by comparing output values produced by the artificial neural network to labeled output values of the network training data set. The second subnetworks are trained by back propagating the error values to modify second parameter values that define the second subnetworks without modifying the first parameter values. The first and second parameter values are stored in a storage component.

Abstract: The disclosed embodiments relate to per-process configuration caches in storage devices. A method is disclosed comprising initiating a new process, the new process associated with a process context; configuring a region in a memory device, the region associated with the process context, wherein the configuring comprises setting one or more cache parameters that modify operation of the memory device; and mapping the process context to the region of the memory device.

Abstract: Systems, apparatuses, and methods of operating memory systems are described. Processing-in-memory capable memory devices are also described, and methods of performing fused-multiply-add operations within the same. Bit positions of bits stored at one or more portions of one or more memory arrays, may be accessed via data lines by activating the same or different access lines. A sensing circuit operatively coupled to a data line may be temporarily formed and measured to determine a state (e.g., a count of the number of bits that are a logic “1”) of accessed bit positions of a data line, and state information may be used to determine a computational result.

Abstract: An apparatus and method for performing memory operations in memory stacks comprising receiving a memory operation request at a first memory controller, where the first memory controller is in included in a first logic die in communication with a first memory die of a first memory technology, from a processor via a first bus. The method further comprising, on a condition that the memory operation request is associated with a second memory technology, communicating the memory operation request to a second memory controller via a side bus, where the second memory controller is included in a second logic die in communication with a second memory die of the second memory technology, and, on a condition that the memory operation request is associated with the first memory technology, performing the memory operation request. The first and second logic dies and the first and second memory dies being stacked on the processor.

Abstract: The present disclosure is directed to systems and methods for a Processing-In-Memory Device that is configured to perform dot product calculations. A sequence control may be used to store data in a memory array according to an allocation pattern. The cells of the memory array may correspond to array elements of the data. The sequence control may apply another array of data to groups of elements within the memory array using the allocation pattern to perform dot product calculations. The dot product calculations may be used, for example, to implement a layer in a convolutional neural network.

Abstract: Methods, systems, and devices related to content-addressable memory for signal development caching are described. In one example, a memory device in accordance with the described techniques may include a memory array, a sense amplifier array, and a signal development cache configured to store signals (e.g., cache signals, signal states) associated with logic states (e.g., memory states) that may be stored at the memory array (e.g., according to various read or write operations). The memory device may also include storage, such as a content-addressable memory, configured to store a mapping between addresses of the signal development cache and addresses of the memory array. In various examples, accessing the memory device may include determining and storing a mapping between addresses of the signal development cache and addresses of the memory array, or determining whether to access the signal development cache or the memory array based on such a mapping.