International Business Machines Corporation

Saiful Islam

Dung Q Nguyen

Sam Gat-Shang Chu

Sam G Chu

David R Terry

Steven J Battle

Maureen Anne Delaney

Eula A Tolentino

Owen Chiang

Jafar Nahidi

Vikas Agarwal

Jose Angel Paredes

Sanjay Dubey

Shelton Siuwah Leung

Joel Abraham Silberman

Stephen Victor Kosonocky

Jae-Joon Kim

Brian William Curran

Mary Douglass Brown

Robert Allen Cordes

Taqi Nasser Buti

Philip George Shephard, Iii

Abraham Mathews

Tuhin Mahmud

Gaurav Mittal

Binglong Zhang

Bjorn P Christensen

Geoffrey Wang

Howard Lawrence Levy

Asit S Ambekar

Zakaria Mahmood Khwaja

Aspects of the invention include obtaining one or more feature values that define an architecture design of a memory array and implementing a machine learning model to obtain a predicted power, performance, and area (PPA) of the memory array based on the one or more features. The predicted PPA output by the machine leaning model is assessed based on predefined PPA goals. A design of an integrated circuit that includes the memory array is finalized and fabricated based on the predicted PPA meeting the predefined PPA goals.

Integrated circuit development using machine learning-based prediction of power, performance, and area

An approach is provided in which a system writes a set of data into a register file entry that includes a first memory array and a second memory array. The register file entry also includes a set of first write ports corresponding to the first memory array and a set of second write ports corresponding to the second memory array. The system configures a selection bit based on determining that a selected one of the set of first write ports is utilized to store the set of data in the first memory array. In turn, the system reads the set of data out of the first memory array based on the configured selection bit.

Implementing write ports in register-file array cell

Reducing power consumption in a multi-slice computer processor that includes a re-order buffer and an architected register file, including: designating an entry in the re-order buffer as being invalid and unwritten; assigning a pending instruction to the entry in the re-order buffer; responsive to assigning the pending instruction to the entry in the re-order buffer, designating the entry as being valid; writing data generated by executing the pending instruction into the re-order buffer; and responsive to writing data generated by executing the pending instruction into the re-order buffer, designating the entry as being written.

Reducing power consumption in a multi-slice computer processor

Methods and apparatus for preventing premature reads from a general purpose register (GPR) including receiving an instruction comprising a source operand identifying a source GPR entry; setting a read-enabled flag based on a value in a particular entry of a source ready vector; if the read-enabled flag indicates data in the source GPR entry is ready for reading, dispatching the received instruction, including performing a read operation of the data in the source GPR entry; and if the read-enabled flag indicates data in the source GPR entry is not ready for reading, dispatching the received instruction without performing a read operation of the data in the source GPR entry.

Preventing premature reads from a general purpose register

A processor having a dependency matrix comprises a first array comprising a plurality of first cells. A second array couples to the first array and comprises a plurality of second cells. A first write port couples to the first array and the second array and writes to the first array and the second array. A first read port couples to the first array and the second array and reads from the first array and the second array. A second read port couples to the first array and reads from the first array. A second write port couples to the second read port, reads from the second read port and writes to the second array.

Dependency matrix with reduced area and power consumption

An integrated circuit (IC) includes power supply interconnects that couple to a power source. The integrated circuit includes electronic devices that perform desired functions and further includes decoupling capacitor circuits that provide noise reduction throughout the integrated circuit. In one embodiment, each decoupling capacitor circuit includes a decoupling capacitor and a switching circuit. The switching circuit connects the decoupling capacitor to the power supply interconnects during a connect mode when the switching circuit detects no substantial decoupling capacitor leakage. However, the switching circuit effectively disconnects the decoupling capacitor from the power supply interconnects during a disconnect mode when the switching circuit detects substantial decoupling capacitor leakage. The decoupling capacitor circuit self-initializes in the connect mode without external control signals and is thus self-contained. Because of the self-contained nature of the decoupling capacitor circuit, an integrated circuit may contain an array of decoupling capacitor circuits without expenditure of substantial chip real estate for respective decoupling capacitor control lines.

Method and apparatus for self-contained automatic decoupling capacitor switch-out in integrated circuits

A register file is often used within integrated circuitry to temporarily hold data. Sometimes this data needs to be retained within the register file for a period of time, such as when there is a stall operation. Conventional register files have utilized a hold multiplexor to perform such a stall operation. The multiplexor however inserts a delay that is undesirable in high performance integrated circuitry. The multiplexor is replaced with a tri-state inverter coupled to the global bit line of the register file, which minimizes this additional delay from the register file data access time.

Register file

A memory circuit includes a global read bit line, a global read bit line latch, and a plurality of sub-arrays, each of which includes first and second local read bit lines, first and second local write bit lines, and first and second pluralities of memory cells interconnected, respectively, with the first and second local read bit lines and the first and second local write bit lines. The local read bit lines are decoupled from the local write bit lines. A local multiplexing block is interconnected with the first and second local read bit lines and is configured to ground the first and second local read bit lines upon assertion of a SLEEP signal, and to selectively interconnect the local read bit lines to the global read bit line. A global multiplexing block is interconnected with the global read bit line and is configured to maintain the global read bit line in a substantially discharged state upon assertion of the SLEEP signal and to interconnect the global read bit line to the global read bit line latch. Also included are design structures for circuits of the kind described.

Memory circuits with reduced leakage power and design structures for same

An apparatus and method are provided for reading a plurality of consecutive entries and writing a plurality of consecutive entries with only one read address and one write address using a 2Read/2Write register file. In one exemplary embodiment, a 64 entry register file array is partitioned into four sub-arrays. Each sub-array contains sixteen entries having one or more 2Read/2Write SRAM cells. The apparatus and method provide a mechanism to write the consecutive entries by only having a 4 to 16 decode of one address. In addition, the apparatus and method provide a mechanism for reading data from the register file array using a starting read word address and two read word lines generated based on the starting read word address. The two read word lines are used to access the two read ports of the entries in the sub-arrays.

Apparatus and method for providing multiple reads/writes using a 2Read/2Write register file array

An energy-saving circuit and method using charge equalization across complementary nodes reduces power consumption in memory circuits and other circuits such as wide multiplexers having complementary high-capacitance nodes. A change detection circuit detects a state change to be applied to the bitlines, and generates a pulse if a state change is to be applied. A pass gate connected between the nodes is activated in response to the pulse to equalize the charge on the bitlines. The driver circuit enable inputs are also delayed, so that the bitlines are not driven until after the charge has been equalized and the pass gate disabled. In one embodiment, the driver circuits are only enabled momentarily by a pulsed output of the change detector and keeper circuits are employed to retain the bitlines in their asserted states.

Energy-saving circuit and method using charge equalization across complementary nodes

A hybrid shift register latch which uses static memory cells for system operations and a dynamic memory cell for testing operations only. An L1 storage element and an L2 storage element are provided in an array cell. The L1 storage element comprises a static random access memory cell. The L1 storage element is used during system and testing operation of the array cell. The L2 storage element comprises a dynamic random access memory cell. The L2 storage element is used only during testing operation of the array cell.

Growing community of inventors

Austin, TX, United States of America

Saiful Islam