Ceremorphic, Inc.

Meta Platforms Technologies, LLC

Robert F Wiser

Neelam Surana

Venkat Mattela

Wei Xiong

Shakti Singh

Jay A Chesavage

Ali Mesgarani

Umanath Ramachandra Kamath

A plurality of integrated circuits each have a central interconnect region which is enclosed by an inner sealring, optional intermediate sealrings, and an outer sealring. Each sealring has a sealring gap for passage of a signal trace which connects a central interconnect of a first integrated circuit to a central interconnect of a second integrated circuit. In first example of the invention, the signal trace remains on a single layer and routes through sealring layer gaps between the first and second IC. In a second example of the invention, vias are used in gaps between sealrings for the signal trace to change layers such that the sealring gaps are not on the same layer. In a third example of the invention, the vias of the second example are replaced by capacitors with plates in adjacent layers.

Chip to chip interconnect beyond sealring boundary

A method may be performed by a leakage monitoring and compensation system configured to estimate and compensate for leakage current in memory unit cells. The method may include identifying a leakage monitoring component associated with a memory unit cell. Further, the method may include sampling multiple leak events during a first exposure window. The method may include storing a first count representing the plurality of leak events sampled during the first exposure window. Each leak event may correspond to a unit of memory leakage. The method may include sampling multiple sensing events during a second exposure window. The method may include detecting a second count representing the sensing events sampled during the second exposure window. The method may include determining a compensation value representing a difference between the first count and the second count.

Method and system for estimating and compensating for leakage current in memory unit cells

A static random access memory (SRAM) has one or more arrays of memory cells, each array of memory cells activated in columns by a wordline. The activated column of memory cells asserts output data onto a plurality of bitlines coupled to output drivers. The SRAM includes a wordline controller generating a variable pulse width wordline which may be reduced sufficient to introduce memory read errors. Each of a high error rate, medium error rate, low error rate, and error-free rate is associated with a pulse width value generated by the wordline controller. A power consumption tradeoff exists between the wordline pulse width and consumed SRAM power. The wordline controller is thereby able to associate a wordline pulse width to an associated error rate for performing tasks which are insensitive to a high error rate or a medium error rate, which are specific to certain neural network training and inference using various NN data types.

Address dependent wordline timing in asynchronous static random access memory

A scan chain architecture with lowered power consumption comprises a multiplexer selecting between a functional input and a test input. The output of the multiplexer is coupled to a low threshold voltage latch and, in test mode, to a standard threshold voltage latch. The low threshold voltage latch and standard threshold voltage latch are configured to store data when a clock input falls, using a master latch functional clock M_F_CLK, master latch test clock M_T_CLK, slave latch functional clock S_F_CLK, and slave latch test clock S_T_CLK. The slave latch has lower power consumption than the master latch.

Process for scan chain in a memory

A static random access memory (SRAM) has one or more arrays of memory cells, each array of memory cells activated in columns by a wordline. The activated column of memory cells asserts output data onto a plurality of bitlines coupled to output drivers. The SRAM includes a wordline controller generating a variable wordline signal pulse width which may be reduced sufficiently to introduce memory read errors. Each of a high error rate, medium error rate, low error rate, and a nearly error-free rate is associated with a pulse width value generated by the wordline controller. A power consumption tradeoff exists between the wordline signal pulse width and consumed SRAM power. The wordline controller is thereby able to associate a wordline signal pulse width with an associated error rate for performing tasks which are insensitive to a high error rate or a medium error rate, which are specific to certain neural network training and inference using various NN data types.

Dynamic adjustment of wordline timing in static random access memory

A memory performing logic functions has two single transistor static ram memory (STSRAM) with drain, source, and gate terminal which can be written, read, and when read, generates an output current. The STSRAMs have drain and source connected in parallel, and when read, generate a current provided to a current comparator amplifier (CCA) which is compared to a reference current Ito generate an output which is at least one of a logical AND, logical NAND, logical OR, logical NOR, or logical exclusive OR (XOR).

One transistor memory bitcell with arithmetic capability

Scan chain for memory with reduced power consumption

A node mesh contains an originating node and several node groups, each node group consisting of one or more nodes with interfaces connected to other nodes of the node group. Each node of a node group has an associated route table with an association between an applied DC voltage and an output interface to couple the input signal to. When the originating node outputs a DC voltage accompanied by differential signaling, each node in turn directs the DC voltage and differential signaling to an output interface as directed by the node local route table to a local termination in a node, which may be coupled to a training processor of inference processor for machine learning applications.

Chip to chip network routing using DC bias and differential signaling

A node mesh contains an originating node coupled to one or more nodes, each node having an communications interface input and a communications interface output. Each node has a route table with an association between a header amplitude and an output interface, such that a header having a particular amplitude causes the input node which received the message to couple the message to an associated communications interface output of the node. When the originating node outputs a message with a header amplitude, each node of the node mesh in turn directs the message to an output interface as directed by the node local route table to a terminating node of the node mesh, where the terminating node may be a training processor or inference processor for machine learning applications.

Growing community of inventors

Santa Cruz, CA, United States of America

Robert F Wiser