Advanced Micro Devices Corporation

Dmitri Yudanov

David Andrew Roberts

Rahul Kumar Agarwal

Milind S Bhagavat

Sergey Blagodurov

Sooraj Puthoor

Brett P Wilkerson

Arkaprava Basu

Mitesh Ramesh Meswani

Michael Ignatowski

Joseph Lee Greathouse

James Michael O'Connor

Bradford Michael Beckmann

Jiasheng Chen

Nicholas P Malaya

An apparatus and method for performing memory operations in memory stacks comprising receiving a memory operation request at a first memory controller, where the first memory controller is in included in a first logic die in communication with a first memory die of a first memory technology, from a processor via a first bus. The method further comprising, on a condition that the memory operation request is associated with a second memory technology, communicating the memory operation request to a second memory controller via a side bus, where the second memory controller is included in a second logic die in communication with a second memory die of the second memory technology, and, on a condition that the memory operation request is associated with the first memory technology, performing the memory operation request. The first and second logic dies and the first and second memory dies being stacked on the processor.

Method and apparatus of integrating memory stacks

A three-dimensional integrated circuit includes a first die structure having a first geometry. The first die structure includes a first region that operates with a first power density and a second region that operates with a second power density. The first power density is less than the second power density. The three-dimensional integrated circuit includes a second die structure having a second geometry. A stacked portion of the second die structure is aligned with the first region. The three-dimensional integrated circuit includes an additional die structure stacked with the first die structure and the second die structure. The additional die structure has the first geometry or the second geometry.

Offset-aligned three-dimensional integrated circuit

An artificial neural network that includes first subnetworks to implement known functions and second subnetworks to implement unknown functions is trained. The first subnetworks are trained separately and in parallel on corresponding known training datasets to determine first parameter values that define the first subnetworks. The first subnetworks are executing on a plurality of processing elements in a processing system. Input values from a network training data set are provided to the artificial neural network including the trained first subnetworks. Error values are generated by comparing output values produced by the artificial neural network to labeled output values of the network training data set. The second subnetworks are trained by back propagating the error values to modify second parameter values that define the second subnetworks without modifying the first parameter values. The first and second parameter values are stored in a storage component.

Concurrent training of functional subnetworks of a neural network

A method and apparatus of integrating memory stacks includes providing a first memory die of a first memory technology and a second memory die of a second memory technology. A first logic die is in communication with the first memory die of the first memory technology, and includes a first memory controller including a first memory control function for interpreting requests in accordance with a first protocol for the first memory technology. A second logic die is in communication with the second memory die of the second memory technology and includes a second memory controller including a second memory control function for interpreting requests in accordance with a second protocol for the second memory technology. A memory operation request is received at the first or second memory controller, and the memory operation request is performed in accordance with the associated first memory protocol or the second memory protocol.

A method for manufacturing a three-dimensional integrated circuit includes attaching a first side of a first die to a first carrier wafer. The method includes preparing a second side of the first die to generate a prepared second side of the first die. The method includes attaching the prepared second side of the first die to a second carrier wafer. The method includes removing the first carrier wafer from the first side of the first die to form a transitional three-dimensional integrated circuit. The method includes attaching a third carrier wafer to a first side of the transitional three-dimensional integrated circuit. The method includes attaching a first side of the second die to a second side of the transitional three-dimensional integrated circuit.

A compute unit configured to execute multiple threads in parallel is presented. The compute unit includes one or more single instruction multiple data (SIMD) units and a fetch and decode logic. The SIMD units have differing numbers of arithmetic logic units (ALUs), such that each SIMD unit can execute a different number of threads. The fetch and decode logic is in communication with each of the SIMD units, and is configured to assign the threads to the SIMD units for execution based on such differing numbers of ALUs.

Heterogeneous graphics processing unit for scheduling thread groups for execution on variable width SIMD units

A modified 1C1T cell detects when the charge in the memory cell drops below a predetermined voltage due to leakage and asserts a refresh signal indicating that refresh needs to be performed on those memory cells associated with the modified 1C1T memory cell. The associated memory cells may be a row, a bank, or other groupings of memory cells. Because temperature affects leakage current, the modified memory cell automatically adjusts for temperature.

Utilizing capacitors integrated with memory devices for charge detection to determine DRAM refresh

A method and processing apparatus for accelerating program processing is provided that includes a plurality of processors configured to process a plurality of tasks of a program and a controller. The controller is configured to determine, from the plurality of tasks being processed by the plurality of processors, a task being processed on a first processor to be a lagging task causing a delay in execution of one or more other tasks of the plurality of tasks. The controller is further configured to provide the determined lagging task to a second processor to be executed by the second processor to accelerate execution of the lagging task.

Multi-processor apparatus and method of detection and acceleration of lagging tasks

A three-dimensional integrated circuit includes a first die having a first geometry. The first die includes a first region that operates with a first power density and a second region that operates with a second power density. The first power density is less than the second power density. The first die includes first electrical contacts disposed in the first region on a first side of the first die along a periphery of the first die. The three-dimensional integrated circuit includes a second die having a second geometry. The second die includes second electrical contacts disposed on a first side of the second die. A stacked portion of the second die is stacked within the periphery of the first die and an overhang portion of the second die extends beyond the periphery of the first die. The second electrical contacts are aligned with and coupled to the first electrical contacts.

Various chip stack power delivery circuits are disclosed. In one aspect, an apparatus is provided that includes a stack of semiconductor chips that has an uppermost semiconductor chip and a lowermost semiconductor chip. A heat spreader is positioned on the uppermost semiconductor chip. A power transfer circuit is configured to transfer electric power from the heat spreader to the uppermost semiconductor chip.

Method and apparatus for power delivery to a die stack via a heat spreader

A technique for accessing memory in an accelerated processing device coupled to stacked memory dies is provided herein. The technique includes receiving a memory access request from an execution unit and identifying whether the memory access request corresponds to memory cells of the stacked dies that are considered local to the execution unit or non-local. For local accesses, the access is made 'directly', that is, without using a bus. A control die coordinates operations for such local accesses, activating particular through-silicon-vias associated with the memory cells that include the data for the access. Non-local accesses are made via a distributed cache fabric and an interconnect bus in the control die. Various other features and details are provided below.

Extreme-bandwidth scalable performance-per-watt GPU architecture

A method and apparatus of performing a memory operation includes receiving a memory operation request at a first memory controller that is in communication with a second memory controller. The first memory controller forwards the memory operation request to the second memory controller. Upon receipt of the memory operation request, the second memory controller provides first information or second information depending on a condition of a pseudo-bank of the second memory controller and a type of the memory operation request.

Method and apparatus of performing a memory operation in a hierarchical memory assembly

A processing device includes a first memory that includes a context buffer. The processing device also includes a processor core to execute threads based on context information stored in registers of the processor core and a memory controller to selectively move a subset of the context information between the context buffer and the registers based on one or more latencies of the threads.

Predicting a context portion to move between a context buffer and registers based on context portions previously used by at least one other thread

A method and apparatus for performing inter-lane power management includes de-energizing one or more execution lanes upon a determination that the one or more execution lanes are to be predicated. Energy from the predicated execution lanes is redistributed to one or more active execution lanes.

Method and apparatus for performing inter-lane power management

A method, a non-transitory computer readable medium, and a processor for repacking dynamic wavefronts during program code execution on a processing unit, each dynamic wavefront including multiple threads are presented. If a branch instruction is detected, a determination is made whether all wavefronts following a same control path in the program code have reached a compaction point, which is the branch instruction. If no branch instruction is detected in executing the program code, a determination is made whether all wavefronts following the same control path have reached a reconvergence point, which is a beginning of a program code segment to be executed by both a taken branch and a not taken branch from a previous branch instruction. The dynamic wavefronts are repacked with all threads that follow the same control path, if all wavefronts following the same control path have reached the branch instruction or the reconvergence point.

Growing community of inventors

Austin, TX, United States of America

Dmitri Yudanov