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-serial 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 the 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: 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 data corresponding to a group of destination neurons. The memory system may, at each time interval of a SNN, pass through data corresponding to a group of pre-synaptic spike events from respective source neurons. The data corresponding to the group of pre-synaptic spike events may be subsequently stored in the memory system.

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: Methods, systems, and devices for error correction for content-addressable memory (CAM) are described. A CAM may store bit vectors as a set of subvectors, which each subvector stored in an independent aspect of the CAM, such as in a separate column or array of memory cells within the CAM. The CAM may similarly segment a queried input bit vector and identify, for each resulting input subvector, whether a matching subvector is stored by the CAM. The CAM may identify a match for the input bit vector when the number of matching subvectors satisfies a threshold. The CAM may validate a match based on comparing a stored bit vector corresponding to the identified match to the input bit vector. The stored bit vector may undergo error correction and may be stored in the CAM or another memory array, such as a dynamic random access memory (DRAM) array.

Abstract: An apparatus (e.g., a content addressable memory system) can have a controller; a first content addressable memory coupled to the controller and a second content addressable memory coupled to the controller. The controller can be configured to cause the first content addressable memory to compare input data to first data stored in the first content addressable memory and cause the second content addressable memory to compare the input data to second data stored in the second content addressable memory such the input data is compared to the first and second data concurrently and replace a result of the comparison of the input data to the first data with a result of the comparison of the input data to the second data in response to determining that the first data is invalid and that the second data corresponds to the first data.

Abstract: A system having multiple devices that can host different versions of an artificial neural network (ANN). In the system, changes to local versions of the ANN can be combined with a master version of the ANN. In the system, a first device can include memory that can store the master version, a second device can include memory that can store a local version of the ANN, and there can be many devices that store local versions of the ANN. The second device (or any other device of the system hosting a local version) can include a processor that can train the local version, and a transceiver that can transmit changes to the local version generated from the training. The first device can include a transceiver that can receive the changes to a local version, and a processing device that can combine the received changes with the master version.

Abstract: A system having multiple devices that can host different versions of an artificial neural network (ANN). In the system, inputs for the ANN can be obfuscated for centralized training of a master version of the ANN at a first computing device. A second computing device in the system includes memory that stores a local version of the ANN and user data for inputting into the local version. The second computing device includes a processor that extracts features from the user data and obfuscates the extracted features to generate obfuscated user data. The second device includes a transceiver that transmits the obfuscated user data. The first computing device includes a memory that stores the master version of the ANN, a transceiver that receives obfuscated user data transmitted from the second computing device, and a processor that trains the master version based on the received obfuscated user data using machine learning.

Abstract: A system having multiple devices that can host different versions of an artificial neural network (ANN) as well as different versions of a feature dictionary. In the system, encoded inputs for the ANN can be decoded by the feature dictionary, which allows for encoded input to be sent to a master version of the ANN over a network instead of an original version of the input which usually includes more data than the encoded input. Thus, by using the feature dictionary for training of a master ANN there can be reduction of data transmission.

Abstract: An apparatus (e.g., a content addressable memory system) can have a controller, a first content addressable memory coupled to the controller, and a second content addressable memory coupled to the controller. The controller can be configured to cause the first content addressable memory to write data in the first content addressable memory, cause the second content addressable memory to write the data in the second content addressable memory, and cause the second content addressable memory to query the data written in the second content addressable memory while the first content addressable memory continues to write the data in the first content addressable memory.

Abstract: Systems, methods and apparatuses to accelerate accessing of borrowed memory over network connection are described. For example, a memory management unit (MMU) of a computing device can be configured to be connected both to the random access memory over a memory bus and to a computer network via a communication device. The computing device can borrow an amount of memory from a remote device over a network connection using the communication device; and applications running in the computing device can use virtual memory addresses mapped to the borrowed memory. When a virtual address mapped to the borrowed memory is used, the MMU translates the virtual address into a physical address and instruct the communication device to access the borrowed memory.

Abstract: Systems, methods and apparatuses of Artificial Neural Network (ANN) applications implemented via Memory as a Service (MaaS) are described. For example, a computing system can include a computing device and a remote device. The computing device can borrow memory from the remote device over a wired or wireless network. Through the borrowed memory, the computing device and the remote device can collaborate with each other in storing an artificial neural network and in processing based on the artificial neural network. Some layers of the artificial neural network can be stored in the memory loaned by the remote device to the computing device. The remote device can perform the computation of the layers stored in the borrowed memory on behalf of the computing device. When the network connection degrades, the computing device can use an alternative module to function as a substitute of the layers stored in the borrowed memory.

Abstract: Systems, methods and apparatuses to provide memory as a service are described. For example, a borrower device is configured to: communicate with a lender device; borrow an amount of memory from the lender device; expand memory capacity of the borrower device for applications running on the borrower device, using at least the local memory of the borrower device and the amount of memory borrowed from the lender device; and service accesses by the applications to memory via communication link between the borrower device and the lender device.

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: Systems, methods and apparatuses of fine grain data migration in using Memory as a Service (MaaS) are described. For example, a memory status map can be used to identify the cache availability of sub-regions (e.g., cache lines) of a borrowed memory region (e.g., a borrowed remote memory page). Before accessing a virtual memory address in a sub-region, the memory status map is checked. If the sub-region has cache availability in the local memory, the memory management unit uses a physical memory address converted from the virtual memory address to make memory access. Otherwise, the sub-region is cached from the borrowed memory region to the local memory, before the physical memory address is used.

Abstract: Systems, methods and apparatuses to throttle network communications for memory as a service are described. For example, a computing device can borrow an amount of random access memory of the lender device over a communication connection between the lender device and the computing device. The computing device can allocate virtual memory to applications running in the computing device, and configure at least a portion of the virtual memory to be hosted on the amount of memory loaned by the lender device to the computing device. The computing device can throttle data communications used by memory regions in accessing the amount of memory over the communication connection according to the criticality levels of the contents stored in the memory regions.

Abstract: Systems, methods and apparatuses to intelligently migrate content involving borrowed memory are described. For example, after the prediction of a time period during which a network connection between computing devices having borrowed memory degrades, the computing devices can make a migration decision for content of a virtual memory address region, based at least in part on a predicted usage of content, a scheduled operation, a predicted operation, a battery level, etc. The migration decision can be made based on a memory usage history, a battery usage history, a location history, etc. using an artificial neural network; and the content migration can be performed by remapping virtual memory regions in the memory maps of the computing devices.

Abstract: Several embodiments of memory devices and systems with selective page-based refresh are disclosed herein. In one embodiment, a memory device includes a controller operably coupled to a main memory having at least one memory region comprising a plurality of memory pages. The controller is configured to track, in one or more refresh schedule tables stored on the memory device and/or on a host device, a subset of memory pages in the plurality of memory pages configured to be refreshed according to a refresh schedule. In some embodiments, the controller is further configured to refresh the subset of memory pages in accordance with the refresh schedule.

Abstract: An apparatus (e.g., a content addressable memory system) can have a controller; a first content addressable memory coupled to the controller and a second content addressable memory coupled to the controller. The controller can be configured to cause the first content addressable memory to compare input data to first data stored in the first content addressable memory and cause the second content addressable memory to compare the input data to second data stored in the second content addressable memory such the input data is compared to the first and second data concurrently and replace a result of the comparison of the input data to the first data with a result of the comparison of the input data to the second data in response to determining that the first data is invalid and that the second data corresponds to the first data.

Abstract: An apparatus (e.g., a content addressable memory system) can have a controller, a first content addressable memory coupled to the controller, and a second content addressable memory coupled to the controller. The controller can be configured to cause the first content addressable memory to write data in the first content addressable memory, cause the second content addressable memory to write the data in the second content addressable memory, and cause the second content addressable memory to query the data written in the second content addressable memory while the first content addressable memory continues to write the data in the first content addressable memory.

Abstract: A method, an apparatus, and a system have been disclosed. An embodiment of the method includes an autonomous memory device receiving a set of instructions, the memory device executing the set of instructions, combining the set of instructions with any data recovered from the memory device in response to the set of instructions into a packet, and transmitting the packet from the memory device.