Abstract: Matrix multiplication process is segregated between two separate dies—a memory die and a compute die to achieve low latency and high bandwidth artificial intelligence (AI) processor. The blocked matrix-multiplication scheme maps computations across multiple processor elements (PE) or matrix-multiplication units. The AI architecture for inference and training includes one or more PEs, where each PE includes memory (e.g., ferroelectric (FE) memory, FE-RAM, SRAM, DRAM, MRAM, etc.) to store weights and input/output I/O data. Each PE also includes a ring or mesh interconnect network to couple the PEs for fast access of information.

Abstract: A device includes, in a first region, a first conductive interconnect, an electrode structure on the first conductive interconnect, where the electrode structure includes a first conductive hydrogen barrier layer and a first conductive fill material. A memory device including a ferroelectric material or a paraelectric material is on the electrode structure. A second dielectric includes an amorphous, greater than 90% film density hydrogen barrier material laterally surrounds the memory device. A via electrode including a second conductive hydrogen barrier material is on at least a portion of the memory device. A second region includes a conductive interconnect structure embedded within a less than 90% film density material.

Abstract: A device includes, in a first region, a first conductive interconnect, an electrode structure on the first conductive interconnect, where the electrode structure includes a first conductive hydrogen barrier layer and a first conductive fill material. A memory device including a ferroelectric material or a paraelectric material is on the electrode structure. A second dielectric includes an amorphous, greater than 90% film density hydrogen barrier material laterally surrounds the memory device. A via electrode including a second conductive hydrogen barrier material is on at least a portion of the memory device. A second region includes a conductive interconnect structure embedded within a less than 90% film density material.

Abstract: A packaging technology to improve performance of an AI processing system resulting in an ultra-high bandwidth 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 can be a first logic die (e.g., a compute chip, CPU, GPU, etc.) while the second die can be a compute chiplet comprising ferroelectric or paraelectric logic. Both dies can include ferroelectric or paraelectric logic. The ferroelectric/paraelectric logic may include AND gates, OR gates, complex gates, majority, minority, and/or threshold gates, sequential logic, etc. The IC package can be in a 3D or 2.5D configuration that implements logic-on-logic stacking configuration. The 3D or 2.5D packaging configurations have chips or chiplets designed to have time distributed or spatially distributed processing. The logic of chips or chiplets is segregated so that one chip in a 3D or 2.5D stacking arrangement is hot at a time.

Abstract: Non lead-based perovskite ferroelectric devices for high density memory and logic applications and methods of fabrication are described. While various embodiments are described with reference to FeRAM, capacitive structures formed herein can be used for any application where a capacitor is desired. For example, the capacitive structure can be used for fabricating ferroelectric based or paraelectric based majority gate, minority gate, and/or threshold gate.

Abstract: A personal digital ID device provides a digital identifier to a service for a predetermined duration in response to user interaction. The user interaction may include a button press. The personal digital ID device may be in the form of a bracelet, a key fob, or other form factor. The service may be provided by a mobile device, in the cloud, or elsewhere.

Abstract: 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.

Abstract: A computer-aided design (CAD) tool is provided for logic optimization and synthesis. The CAD tool executes a process that involves optimizing power, performance, and area (PPA) of a logic circuit by minimizing a number of CMOS gates, and majority and/or minority gates in the circuit and its depth. The CAD tool implements a methodology of optimizing logic synthesis based on a mix of standard cell libraries (such as AND, OR, NAND, NOR, XOR, Multiplexer, full adder, half adder, etc.) and varying input majority and minority gates (where the number of inputs in the minority and majority gates could vary as odd numbers from 3 and above). The standard cell libraries cells may contain minority and/or majority gates.

Abstract: Asynchronous circuits implemented using threshold gate(s) and/or majority gate(s) (or minority gate(s)) are described. The new class of asynchronous circuits can operate at lower power supply levels (e.g., less than 1 V on advanced technology nodes) because stack of devices between a supply node and ground are significantly reduced compared to traditional asynchronous circuits. The asynchronous circuits here result in area reduction (e.g., 3× reduction compared to traditional asynchronous circuits) and provide higher throughput/mm2 (e.g., 2× higher throughput compared to traditional asynchronous circuits). The threshold gate(s), majority/minority gate(s) can be implemented using capacitive input circuits. The capacitors can have linear dielectric or non-linear polar material as dielectric.

Abstract: A memory is provided which comprises a capacitor including non-linear polar material. The capacitor may have a first terminal coupled to a node (e.g., a storage node) and a second terminal coupled to a plate-line. The capacitors can be a planar capacitor or non-planar capacitor (also known as pillar capacitor). The memory includes a transistor coupled to the node and a bit-line, wherein the transistor is controllable by a word-line, wherein the plate-line is parallel to the bit-line. The memory includes a refresh circuitry to refresh charge on the capacitor periodically or at a predetermined time. The refresh circuit can utilize one or more of the endurance mechanisms. When the plate-line is parallel to the bit-line, a specific read and write scheme may be used to reduce the disturb voltage for unselected bit-cells. A different scheme is used when the plate-line is parallel to the word-line.

Abstract: To compensate switching of a dielectric component of a non-linear polar material based capacitor, an explicit dielectric capacitor is added to a memory bit-cell and controlled by a signal opposite to the signal driven on a plate-line.

Abstract: Asynchronous circuit elements are described. Asynchronous circuit elements include a consensus element (c-element), completion tree, and validity tree. The c-element is implemented using adjustable threshold based multi-input capacitive circuitries. The completion tree comprises a plurality of c-elements organized in a tree formation. The validity tree comprises OR gates followed by c-elements. The multi-input capacitive circuitries include capacitive structures that may comprise linear dielectric, paraelectric dielectric, or ferroelectric dielectric. The capacitors can be planar or non-planar. The capacitors may be stacked vertically to reduce footprint of the various asynchronous circuitries.

Abstract: A computer-aided design (CAD) tool is provided for logic optimization and synthesis. The CAD tool executes a process that involves optimizing power, performance, and area (PPA) of a logic circuit by minimizing a number of CMOS gates, and majority and/or minority gates in the circuit and its depth. The CAD tool implements a methodology of optimizing logic synthesis based on a mix of standard cell libraries (such as AND, OR, NAND, NOR, XOR, Multiplexer, full adder, half adder, etc.) and varying input majority and minority gates (where the number of inputs in the minority and majority gates could vary as odd numbers from 3 and above). The standard cell libraries cells may contain minority and/or majority gates.

Abstract: A high-density low voltage ferroelectric (or paraelectric) memory bit-cell that includes a planar ferroelectric or paraelectric capacitor. The memory bit-cell comprises 1T1C configuration, where a plate-line is parallel to a word-line, or the plate-line is parallel to a bit-line. The memory bit-cell can be 1TnC, where ‘n’ is a number. In a 1TnC bit-cell, the capacitors are vertically stacked allowing for multiple values to be stored in a single bit-cell. The memory bit-cell can be multi-element FE gain bit-cell. In a multi-element FE gain bit-cell, data sensing is done with signal amplified by a gain transistor in the bit-cell. As such, higher storage density is realized using multi-element FE gain bit-cells. In some examples, the 1T1C, 1TnC, and multi-element FE gain bit-cells are multi-level bit-cells. To realize multi-level bit-cells, the capacitor is placed in a partially switched polarization state by applying different voltage levels or different time pulse widths at the same voltage level.

Abstract: To compensate switching of a dielectric component of a non-linear polar material based capacitor, an explicit dielectric capacitor is added to a memory bit-cell and controlled by a signal opposite to the signal driven on a plate-line.

Abstract: A packaging technology to improve performance of an AI processing system resulting in an ultra-high bandwidth 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 can be a first logic die (e.g., a compute chip, CPU, GPU, etc.) while the second die can be a compute chiplet comprising ferroelectric or paraelectric logic. Both dies can include ferroelectric or paraelectric logic. The ferroelectric/paraelectric logic may include AND gates, OR gates, complex gates, majority, minority, and/or threshold gates, sequential logic, etc. The IC package can be in a 3D or 2.5D configuration that implements logic-on-logic stacking configuration. The 3D or 2.5D packaging configurations have chips or chiplets designed to have time distributed or spatially distributed processing. The logic of chips or chiplets is segregated so that one chip in a 3D or 2.5D stacking arrangement is hot at a time.

Abstract: A memory is provided which comprises a capacitor including non-linear polar material. The capacitor may have a first terminal coupled to a node (e.g., a storage node) and a second terminal coupled to a plate-line. The capacitors can be a planar capacitor or non-planar capacitor (also known as pillar capacitor). The memory includes a transistor coupled to the node and a bit-line, wherein the transistor is controllable by a word-line, wherein the plate-line is parallel to the bit-line. The memory includes a refresh circuitry to refresh charge on the capacitor periodically or at a predetermined time. The refresh circuit can utilize one or more of the endurance mechanisms. When the plate-line is parallel to the bit-line, a specific read and write scheme may be used to reduce the disturb voltage for unselected bit-cells. A different scheme is used when the plate-line is parallel to the word-line.

Abstract: A process integration and patterning flow used to pattern a memory array area for an embedded memory without perturbing a fabricating process for logic circuitries. The fabrication process uses a pocket mask (e.g., a hard mask) to decouple the etching process of a memory array area and non-memory area. Such decoupling allows for a simpler fabrication process with little to no impact on the current fabrication process. The fabrication process may use multiple pocket masks to decouple the etching process of the memory array area and the non-memory area. This fabrication process (using multiple pocket masks) allows to avoid exposure of memory material into a second pocket etch chamber. The process of etching memory material is decoupled from the process of etching an encapsulation material. Examples of embedded memory include dynamic random-access memory and ferroelectric random-access memory.

Abstract: To compensate switching of a dielectric component of a non-linear polar material based capacitor, an explicit dielectric capacitor is added to a memory bit-cell and controlled by a signal opposite to the signal driven on a plate-line.

Abstract: 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.