Described is a packaging technology to improve performance of an AI processing system. An IC package is provided which comprises: a substrate; a first die on the substrate, and a second die stacked over the first die. The first die includes memory and the second die includes computational logic. The first die comprises a ferroelectric RAM (FeRAM) having bit-cells. Each bit-cell comprises an access transistor and a capacitor including ferroelectric material. The access transistor is coupled to the ferroelectric material. The FeRAM can be FeDRAM or FeSRAM. The memory of the first die may store input data and weight factors. The computational logic of the second die is coupled to the memory of the first die. The second die is an inference die that applies fixed weights for a trained model to an input data to generate an output. In one example, the second die is a training die that enables learning of the weights.

An adaptive memory management and control circuitry (AMMC) to provide extended test, performance, and power optimizing capabilities for a resistive memory is disclosed herein. In one embodiment, a resistive memory comprises a resistive memory array and an Adaptive Memory Management and Control circuitry (AMMC) that is coupled to the resistive memory array. The AMMC is configured with extended test, reliability, performance and power optimizing capabilities for the resistive memory.

Systems, methods and devices are disclosed for a smart compute memory circuitry that has the flexibility to perform a wide range of functions inside the memory via logic circuitry and an integrated processor. In one embodiment, the smart compute memory circuitry comprises an integrated processor and logic circuitry to enable adaptive System on a Chip (SOC) and electronics subsystem power or performance improvements, and adaptive memory management and control for the smart compute memory circuitry. A resistive memory array is coupled to the integrated processor.

Described is a packaging technology to improve performance of an AI processing system. An IC package is provided which comprises: a substrate; a first die on the substrate, and a second die stacked over the first die. The first die includes memory and the second die includes computational logic. The first die comprises a ferroelectric RAM (FeRAM) having bit-cells. Each bit-cell comprises an access transistor and a capacitor including ferroelectric material. The access transistor is coupled to the ferroelectric material. The FeRAM can be FeDRAM or FeSRAM. The memory of the first die may store input data and weight factors. The computational logic of the second die is coupled to the memory of the first die. The second die is an inference die that applies fixed weights for a trained model to an input data to generate an output. In one example, the second die is a training die that enables learning of the weights.

Provided are a weight matrix circuit and a weight matrix input circuit. The weight matrix circuit includes a memory array including n input lines, m output lines, and n×m resistive memory devices each connected to the n input lines and the m output lines and each having a non-linear current-voltage characteristic, an input circuit connected to each of the input lines, and an output circuit connected to each of the output lines. The input circuit is connected to the resistive memory devices such that the weight matrix circuit has a linear current-voltage characteristic.

Endurance mechanisms are introduced for memories such as non-volatile memories for broad usage including caches, last-level cache(s), embedded memory, embedded cache, scratchpads, main memory, and storage devices. Here, non-volatile memories (NVMs) include magnetic random-access memory (MRAM), resistive RAM (ReRAM), ferroelectric RAM (FeRAM), phase-change memory (PCM), etc. In some cases, features of endurance mechanisms (e.g., randomizing mechanisms) are applicable to volatile memories such as static random-access memory (SRAM), and dynamic random-access memory (DRAM). The endurance mechanisms include a wear leveling scheme that uses index rotation, outlier compensation to handle weak bits, and random swap injection to mitigate wear out attacks.

Methods, systems, and devices for refresh command management are described. A memory device may conduct a refresh operation to preserve the integrity of data stored to one or more memory cells. In some examples, the frequency of refresh operations conducted may be based on the memory device's temperature and may be initiated based on one or more commands received from an external device (e.g., a host device). Each command may be transmitted by the host device at a defined rate, which may impact the rate at which the memory device conducts one or more refresh operations. The memory device may postpone or skip at least a portion of one or more refresh operations based on one or more operating parameters of the memory device.

A voltage of a conductive line, such as a control line, a data line, or a voltage supply line associated with a memory die may be monitored. A frequency response of the voltage may be analyzed to determine if the conductive line may be operating at or near a specific frequency, such as a resonance frequency. If the conductive line is operating at or near the specific frequency, an action, such as a memory operation, may be performed to mitigate the resonance of the conductive line. The monitoring, analyzing, and action performing may be accomplished with circuitry of the memory die.

Exemplary apparatus includes a nonvolatile memory, a volatile memory separate from the nonvolatile memory, and a controller configured to access the volatile memory and the nonvolatile memory. Exemplary volatile memory is configured to function as a read/write cache. The controller may be configured to perform a read/modify/write memory operation that involves both the volatile memory and the nonvolatile memory. Exemplary devices may have a host interface and may include a data connection configured to perform double data rate data transfer. Exemplary volatile memory may support byte-granularity memory read operations, and the density of the volatile memory may be substantially less than the density of the nonvolatile memory.

Described is a packaging technology to improve performance of an AI processing system. An IC package is provided which comprises: a substrate; a first die on the substrate, and a second die stacked over the first die. The first die includes memory and the second die includes computational logic. The first die comprises a ferroelectric RAM (FeRAM) having bit-cells. Each bit-cell comprises an access transistor and a capacitor including ferroelectric material. The access transistor is coupled to the ferroelectric material. The FeRAM can be FeDRAM or FeSRAM. The memory of the first die may store input data and weight factors. The computational logic of the second die is coupled to the memory of the first die. The second die is an inference die that applies fixed weights for a trained model to an input data to generate an output. In one example, the second die is a training die that enables learning of the weights.