Abstract: Embodiments of the present disclosure generally relate to electronic devices, and more specifically, to multi-level phase change devices. In one embodiment, a memory cell device is provided. The memory cell device generally includes a top surface, a bottom surface and a cell body between the top surface and the bottom surface. The cell body may include a plurality of phase change material layers, which may be used to store data of the cell. In another embodiment, a method of programming a memory cell is provided. The method generally may include applying a sequence of different pulses to each phase change material layer of the cell as the voltage of each pulse in the sequence is ratcheted down from the start of a write cycle to the end of a write cycle.

Abstract: A method of computer processing is disclosed comprising receiving a data packet at a processing node of a neural network, performing a calculation of the data packet at the processing node to create a processed data packet, attaching a tag to the processed data packet, transmitting the processed data packet from the processing node to a receiving node during a systolic pulse, receiving the processed data packet at the receiving node, performing a clockwise convolution on the processed data packet and a counter clockwise convolution on the processed data packet, performing an adding function and backpropagating results of the performed sigmoid function to each of the processing nodes that originally processed the data packet.

Abstract: Some embodiments include a special-purpose hardware accelerator that can perform specialized machine learning tasks during both training and inference stages. For example, this hardware accelerator uses a systolic array having a number of data processing units (“DPUs”) that are each connected to a small number of other DPUs in a local region. Data from the many nodes of a neural network is pulsed through these DPUs with associated tags that identify where such data was originated or processed, such that each DPU has knowledge of where incoming data originated and thus is able to compute the data as specified by the architecture of the neural network. These tags enable the systolic neural network engine to perform computations during backpropagation, such that the systolic neural network engine is able to support training.

Abstract: Some embodiments include a special-purpose hardware accelerator that can perform specialized machine learning tasks during both training and inference stages. For example, this hardware accelerator uses a systolic array having a number of data processing units (“DPUs”) that are each connected to a small number of other DPUs in a local region. Data from the many nodes of a neural network is pulsed through these DPUs with associated tags that identify where such data was originated or processed, such that each DPU has knowledge of where incoming data originated and thus is able to compute the data as specified by the architecture of the neural network. These tags enable the systolic neural network engine to perform computations during backpropagation, such that the systolic neural network engine is able to support training.

Abstract: A method of computer processing is disclosed comprising receiving a data packet at a processing node of a neural network, performing a calculation of the data packet at the processing node to create a processed data packet, attaching a tag to the processed data packet, transmitting the processed data packet from the processing node to a receiving node during a systolic pulse, receiving the processed data packet at the receiving node, performing a clockwise convolution on the processed data packet and a counter clockwise convolution on the processed data packet, performing an adding function and backpropagating results of the performed sigmoid function to each of the processing nodes that originally processed the data packet.

Abstract: Some embodiments include a special-purpose hardware accelerator that can perform specialized machine learning tasks during both training and inference stages. For example, this hardware accelerator uses a systolic array having a number of data processing units (“DPUs”) that are each connected to a small number of other DPUs in a local region. Data from the many nodes of a neural network is pulsed through these DPUs with associated tags that identify where such data was originated or processed, such that each DPU has knowledge of where incoming data originated and thus is able to compute the data as specified by the architecture of the neural network. These tags enable the systolic neural network engine to perform computations during backpropagation, such that the systolic neural network engine is able to support training.

Abstract: Some embodiments include a special-purpose hardware accelerator that can perform specialized machine learning tasks during both training and inference stages. For example, this hardware accelerator uses a systolic array having a number of data processing units (“DPUs”) that are each connected to a small number of other DPUs in a local region. Data from the many nodes of a neural network is pulsed through these DPUs with associated tags that identify where such data was originated or processed, such that each DPU has knowledge of where incoming data originated and thus is able to compute the data as specified by the architecture of the neural network. These tags enable the systolic neural network engine to perform computations during backpropagation, such that the systolic neural network engine is able to support training.

Abstract: A method of computer processing is disclosed comprising receiving a data packet at a processing node of a neural network, performing a calculation of the data packet at the processing node to create a processed data packet, attaching a tag to the processed data packet, transmitting the processed data packet from the processing node to a receiving node during a systolic pulse, receiving the processed data packet at the receiving node, performing a clockwise convolution on the processed data packet and a counter clockwise convolution on the processed data packet, performing an adding function and backpropagating results of the performed sigmoid function to each of the processing nodes that originally processed the data packet.

Abstract: A method of computer processing is disclosed comprising receiving a data packet at a processing node of a neural network, performing a calculation of the data packet at the processing node to create a processed data packet, attaching a tag to the processed data packet, transmitting the processed data packet from the processing node to a receiving node during a systolic pulse, receiving the processed data packet at the receiving node, performing a clockwise convolution on the processed data packet and a counter clockwise convolution on the processed data packet, performing an adding function and backpropagating results of the performed sigmoid function to each of the processing nodes that originally processed the data packet.

Abstract: A method of computer processing is disclosed comprising receiving a data packet at a processing node of a neural network, performing a calculation of the data packet at the processing node to create a processed data packet, attaching a tag to the processed data packet, transmitting the processed data packet from the processing node to a receiving node during a systolic pulse, receiving the processed data packet at the receiving node, performing a clockwise convolution on the processed data packet and a counter clockwise convolution on the processed data packet, performing an adding function and backpropagating results of the performed sigmoid function to each of the processing nodes that originally processed the data packet.

Abstract: A method of computer processing is disclosed comprising receiving a data packet at a processing node of a neural network, performing a calculation of the data packet at the processing node to create a processed data packet, attaching a tag to the processed data packet, transmitting the processed data packet from the processing node to a receiving node during a systolic pulse, receiving the processed data packet at the receiving node, performing a clockwise convolution on the processed data packet and a counter clockwise convolution on the processed data packet, performing an adding function and backpropagating results of the performed sigmoid function to each of the processing nodes that originally processed the data packet.

Abstract: A magnetic memory array and a method for implementing the magnetic memory array for use in Solid-State Drives (SSDs) are provided. A plurality of magnetic pillar memory cells is formed using a deposition and/or growth process to produce a magnetic memory array substantially avoiding milling of magnetic materials.

Abstract: A non-volatile storage apparatus is proposed that includes a plurality of serially connected non-volatile reversible resistance-switching memory cells, a plurality of word lines such that each of the memory cells is connected to a different word line, a bit line connected to a first end of the serially connected memory cells and a switch connected to a second end of the serially connected memory cells. In one embodiment, the memory cells include a reversible resistance-switching structure comprising a first material, a second material and a reversible resistance-switching interface between the first material and the second material, a channel, and means for switching current between current flowing through the channel and current flowing through the reversible resistance-switching interface in order to program and read the reversible resistance-switching interface. A process for manufacturing the memory is also disclosed.

Abstract: Various aspects directed towards facilitating error management within a shared non-volatile memory (NVM) architecture are disclosed. Data is stored in an NVM array, and error correction vector (ECV) information associated with the NVM array is stored in a content addressable memory (CAM). A parallel query of the NVM array and the CAM is then performed, which includes a query of the NVM array that yields a readout of the NVM array, and a query of the CAM that yields an ECV corresponding to the readout of the NVM array.

Abstract: A non-volatile storage apparatus is proposed that includes a plurality of serially connected non-volatile reversible resistance-switching memory cells, a plurality of word lines such that each of the memory cells is connected to a different word line, a bit line connected to a first end of the serially connected memory cells and a switch connected to a second end of the serially connected memory cells. In one embodiment, the memory cells include a reversible resistance-switching structure comprising a first material, a second material and a reversible resistance-switching interface between the first material and the second material, a channel, and means for switching current between current flowing through the channel and current flowing through the reversible resistance-switching interface in order to program and read the reversible resistance-switching interface. A process for manufacturing the memory is also disclosed.

Abstract: Embodiments of the present disclosure generally relate to non-volatile memory and, in particular, non-volatile memory with adjustable cell bit shapes. In one embodiment, an adjustable memory cell is provided. The memory cell generally includes a gate electrode, at least one recording layer and a channel layer. The channel layer generally is capable of supporting a depletion region and is disposed between the gate electrode and the at least one recording layer. In this embodiment, upon activating the gate, the channel layer may be depleted and current initially flowing through the channel may be steered through the at least one recording layer.

Abstract: Embodiments of the present disclosure generally relate to a cloud computing network, or datacenter network, and a method of transferring information among processing nodes in a cloud computing network or datacenter. The network may include a hub that is coupled to a plurality of nodes so that data is transferred between nodes through the hub. Data from different nodes may be written into a slot within the hub, read, and then written into a slot within the destination node. Due to the proximity of the nodes to the hub, or even due to the amount of data to be written, the data may be written at different clock phases. The read may occur one or more clock cycles after the data has been written into the hub.

Abstract: A non-volatile storage apparatus is proposed that includes a plurality of serially connected non-volatile reversible resistance-switching memory cells, a plurality of word lines such that each of the memory cells is connected to a different word line, a bit line connected to a first end of the serially connected memory cells and a switch connected to a second end of the serially connected memory cells. In one embodiment, the memory cells include a reversible resistance-switching structure comprising a first material, a second material and a reversible resistance-switching interface between the first material and the second material, a channel, and means for switching current between current flowing through the channel and current flowing through the reversible resistance-switching interface in order to program and read the reversible resistance-switching interface. A process for manufacturing the memory is also disclosed.

Abstract: A non-volatile storage apparatus is proposed that includes a plurality of serially connected non-volatile reversible resistance-switching memory cells, a plurality of word lines such that each of the memory cells is connected to a different word line, a bit line connected to a first end of the serially connected memory cells and a switch connected to a second end of the serially connected memory cells. In one embodiment, the memory cells include a reversible resistance-switching structure comprising a first material, a second material and a reversible resistance-switching interface between the first material and the second material, a channel, and means for switching current between current flowing through the channel and current flowing through the reversible resistance-switching interface in order to program and read the reversible resistance-switching interface. A process for manufacturing the memory is also disclosed.

Abstract: A non-volatile storage apparatus is proposed that includes a plurality of serially connected non-volatile reversible resistance-switching memory cells, a plurality of word lines such that each of the memory cells is connected to a different word line, a bit line connected to a first end of the serially connected memory cells and a switch connected to a second end of the serially connected memory cells. In one embodiment, the memory cells include a reversible resistance-switching structure comprising a first material, a second material and a reversible resistance-switching interface between the first material and the second material, a channel, and means for switching current between current flowing through the channel and current flowing through the reversible resistance-switching interface in order to program and read the reversible resistance-switching interface. A process for manufacturing the memory is also disclosed.