International Business Machines Corporation

Peter Altevogt

Thomas Pflueger

Cedric Lichtenau

Wolfgang Gellerich

Martin Bernhard Schmidt

Juergen Pille

Gero Dittmann

Andreea Anghel

Cédric Lichtenau

Raiko Nitzsche

Silvio Wiedrich

Malcolm Scott Ware

Christoph Raisch

Charles Robert Lefurgy

Harry S Barowski

Wesley Michael Felter

Marcus Felipe Fontoura

Martin Schwidefsky

Jason Yeong Zien

Hans Boettiger

Lutz Stiege

Silvio Weidrich

The disclosure relates to an initialization circuit for initializing memory cells of a memory array including a common bit line. Individual memory cells are coupled to the common bit line of the memory array via at least one pass element of the individual memory cells. The initialization circuit is operable for receiving a set of partition addresses specifying the partitions, i.e. the memory cells to be initialized. The initialization circuit is operable for successively initializing one cell of the partitions to be initialized and iteratively initializing the remaining memory cells of the partitions to be initialized. A number of memory cells initialized simultaneously in one iteration increases from one iteration to another iteration. Initializing a certain memory cell comprises activating the pass element of the cell so that the memory cell is connected to the bit line. Further aspects relate to methods for initializing memory cells and semiconductor circuits.

Erasing a partition of an SRAM array with hardware support

The disclosure relates to an initialization circuit for initializing memory cells of a memory array. The individual memory cells are coupled to a common bit line of the memory array via at least one pass element of the individual memory cells. Each individual memory cell comprises a charge-based storage element including a capacitance. The initialization circuit activates the pass elements of a plurality of the memory cells to be initialized such that the capacitances of the plurality of memory cells are connected simultaneously to the common bit line. Further, aspects of the disclosure relate to a method for initializing memory cells and a semiconductor circuit.

Semiconductor circuit including an initialization circuit for initializing memory cells and clearing of relatively large blocks of memory

Embodiments of the invention include a computer-implemented method of processor branch prediction. This method aims at training a machine-learning model of processor branch behavior while a processing unit executes computer instructions. Such instructions include branch instructions, load instructions and store instructions. The load instructions and the store instructions cause a control unit of the processing unit to load data from a memory into processor registers and store data from the processor registers to the memory, respectively. Basically, the training of the model involves, for each of N branch instructions (N&#x3e;2) encountered whilst the processing unit executes said branch instructions: identifying a next branch instruction; and feeding the machine-learning model with carefully chosen inputs.

Computerized branch predictions and decisions

Embodiments of memory block erasure are described herein. An aspect includes determining an initial word line set consisting of a single word line. Another aspect includes activating the single word line such that a first memory cell that is connected to the single word line is erased by the activation. Another aspect includes determining a first word line set consisting of the single word line and one additional word line, and wherein the one additional word line corresponds to a second memory cell have a maximum distance from the first memory cell along a bit line that includes the first memory cell and the second memory cell. Another aspect includes activating the first word line set, such that a respective memory cell that is connected to each of the first word line set is erased by the activation.

Memory block erasure

A processor unit for processing storage instructions. The processor unit comprises a detection logic unit configured to identify at least two storage instructions for moving addressable words between registers of the processor unit and neighboring storage locations. The processor unit further comprises a combination unit configured to combine the identified instructions into a single combined instruction; and a data movement unit configured to move the words using the combined instruction.

Method and apparatus for processing storage instructions

The present disclosure relates to storing data in a computer system. The computer system comprising a main memory coupled to a processor and a cache hierarchy. The main memory comprises a predefined bit pattern replacing existing data of the main memory. Aspects include storing the predefined bit pattern into a reference storage of the computer system. At least one bit in a cache directory entry of a first cache line of the cache hierarchy can be set. Upon receiving a request to read the content of the first cache line, the request can be redirected to the predefined bit pattern in the reference storage based on the value of the set bit of the first cache line.

Storing data into a memory

A method for dynamically selecting a size of a memory access may be provided. The method comprises accessing blocks having a variable number of consecutive cache lines, maintaining a vector with entries of past utilizations for each block size, and adapting said block size before a next access to the blocks.

Selection of variable memory-access size

The present disclosure relates to a method for instruction processing with a processor having multiple execution units. The processor includes a dependency cache containing instructions in association with respective execution unit indicators. The method includes: tracking the number of dependent instructions currently assigned to each execution unit of the processor respectively. In response to receiving an instruction of a dependency chain, the execution unit assigned to a previous instruction of the dependency chain on which depends the received instruction may be identified in the dependency cache. In case more than a predefined maximum number of dependent instructions of at least one dependency chain is currently assigned to the identified execution unit, another execution unit of the processor may be selected for scheduling the received instruction, otherwise the received instruction may be scheduled on the identified execution unit.

Enhanced performance-aware instruction scheduling

Embodiments of memory block erasure are described herein. An aspect includes determining a first word line set consisting of a first plurality of word lines. Another aspect includes activating the first plurality of word lines, such that a respective memory cell that is connected to each of the first plurality of word lines is erased by the activation of the first plurality of word lines. Another aspect includes determining a second word line set, wherein the second word line set consists of the first word line set and a second plurality of word lines. Another aspect includes simultaneously activating the first plurality of word lines and the second plurality of word lines, such that a respective memory cell that is connected to each of the second plurality of word lines is erased by the activation of the first plurality of word lines and the second plurality of word lines.

A processor with multiple execution units for instruction processing is provided. The processor comprises an instruction decode and issue logic and a control logic for resolving register access conflicts between subsequent instructions and a dependency cache, which comprises a receiving logic for receiving an execution unit indicator indicative of the execution unit the instruction is planned to be executed on, a storing logic responsive to the receiving logic for storing the received execution unit indicator, and a retrieving logic responsive to a request from the instruction decode and issue logic for providing the stored execution unit indicator for an instruction. The instruction decode and issue logic is adapted for requesting execution unit indicator for an instruction from the dependency cache and to assign the instruction to one respective of the execution units dependent on the execution unit indicator received from the dependency cache.

Performance-aware instruction scheduling

A method of prefetching data is provided including monitoring sequences of memory addresses of data being accessed by a system, whereby sequences of m+1 memory addresses each are continually identified; and for each identified sequence: converting, upon identifying said each sequence, memory addresses of said each sequence into m relative addresses, whereby each of the m relative addresses is relative to a previous memory address in said each sequence, so as to obtain an auxiliary sequence of m relative addresses; upon converting said memory addresses, feeding said auxiliary sequence of m relative addresses as input to a trained machine learning model for it to predict p relative addresses of next memory accesses by the system, where p&#x2265;1; and prefetching data at memory locations associated with one or more memory addresses that respectively correspond to one or more of the p relative addresses predicted.

Computerized methods for prefetching data based on machine learned sequences of memory addresses

The present invention is a system for enhancing the decoding performance of text indexes of an indexed collection of text documents. The posting list includes a plurality of list entries being of variable size depending on the value stored in each posting list entry. The inventive system is based on a statistical analysis of the posting list entries of the plurality of list entries. Prior to decoding, the system can select a decoding routine being appropriate to decode the most frequent type of posting list entries or of a plurality of various posting list entries. After selecting a particular decoding routine and executing the decoding routine for decoding of the entire posting list, a failure check is performed. In the exceptional case that a failure has occurred, a second alternative decoding routine is selected and executed.

Growing community of inventors

Ettlingen, Germany

Peter Altevogt