International Business Machines Corporation

Mark David Bellows

Tolga Ozguner

Ryan Abel Heckendorf

Paul Allen Ganfield

Kent Harold Haselhorst

Chad Balling McBride

John David Irish

Mark S Fredrickson

David Alan Norgaard

Steven Paul Jones

Lonny Lambrecht

Scott Douglas Frei

Melissa Ann Barnum

Brian M McKevett

Ryan Abel Heakendorf

A first first-in-first-out (FIFO) memory may receive first processor input from a first processor group that includes a first processor. The first processor group is configured to execute program code based on the first processor input that includes a set of input signals, a clock signal, and corresponding data. The first FIFO may store the first processor input and may output the first processor input to a second FIFO memory and to a second processor according to a first delay. The second FIFO memory may store the first processor input and may output the first processor input to a third processor according to a second delay. The second processor may execute at least a first portion of the program code and the third processor may execute at least a second portion of the program code responsive to the first processor input.

Delayed execution of program code on multiple processors

Methods and systems with delayed execution of multiple processors

A method, an apparatus, and a computer program are provided to account for data stored in Dynamic Random Access Memory (DRAM) write buffers. There is difficulty in tracking the data stored in DRAM write buffers. To alleviate the difficulty, a cache line list is employed. The cache line list is maintained in a memory controller, which is updated with data movement. This list allows for ease of maintenance of data without loss of consistency.

Memory controller to utilize DRAM write buffers

A mechanism is provided to reuse functional data buffers. With Extreme Data Rate (XDR&#x2122;) Dynamic Random Access Memory (DRAM), test patterns are employed to dynamically calibrate data with the clock. To perform this task, data buffers are employed to store data and commands for the calibration patterns. However, there are different procedures and requirements for transmission and reception calibrations. Hence, to reduce the amount of hardware needed to perform transmission and reception calibrations, the data buffers employ additional front end circuitry to reuse the buffers for both tasks.

Reuse of functional data buffers for pattern buffers in XDR DRAM

In an aspect, a first method of interfacing a processor and memory is provided. The first method includes the steps of (1) generating a first command and associated address of a first type that does not indicate a rank of memory targeted by the first command; (2) determining whether the memory includes a plurality of ranks; (3) if the memory includes a plurality of ranks, employing the processor to update the address associated with the first command to indicate a memory rank targeted by the first command; (4) if the memory does not include a plurality of ranks, employing the processor to update the address associated with the first command to indicate the memory does not include a plurality of ranks; and (5) converting the first command and associated updated address to a second command and associated address that are employed to access the memory. Numerous other aspects are provided.

Rank select operation between an XIO interface and a double data rate interface

The present invention generally relates to memory controllers operating in a system containing a variable system clock. The memory controller may exchange data with a processor operating at a variable processor clock frequency. However the memory controller may perform memory accesses at a constant memory clock frequency. Asynchronous buffers may be provided to transfer data across the variable and constant clock domains. To prevent read buffer overflow while switching to a lower processor clock frequency, the memory controller may quiesce the memory sequencers and pace read data from the sequencers at a slower rate. To prevent write data under runs, the memory controller's data flow logic may perform handshaking to ensure that write data is completely received in the buffer before performing a write access.

Memory controller operating in a system with a variable system clock

A command translation method, apparatus and system are provided for interfacing a processor and a memory. The processor initiates a memory system command in an extreme data rate (XDR) command format which is automatically converted by the command translation method, apparatus and system into a memory system command in a double data rate (DDR) format for forwarding to the memory. Associated with converting the memory system command to the DDR command format is controlling timing of one or more signals presented to the memory interface, the one or more signals being associated with processing the memory system command in the DDR command format. The processor has associated therewith an XDR memory interface controller which adjusts one or more timing parameters of the memory system command in the XDR command format so that DDR timing requirements for the memory system command in the DDR command format are met.

Memory command and address conversion between an XDR interface and a double data rate interface

Managing write-to-read turnarounds in an early read after write memory system is presented. Memory controller logic identifies a write operation's bank set, allows a different bank set read operation to issue prior to the write operation's completion, and allows a same bank set read operation to issue once the write operation completes. The memory controller includes operation counter logic, operation selection logic, operation acceptance logic, command formatting logic, and memory interface logic. The operation counter logic receives new-operation-related signals from the operation acceptance logic and, in turn, provides signals to the operation selection logic and the operation acceptance logic as to when to issue a read operation that corresponds to either an even DRAM bank or an odd DRAM bank.

Managing write-to-read turnarounds in an early read after write memory system

Embodiments of the present invention optimize data bandwidth across an asynchronous buffer in a system with a variable clock domain. A move signal may be asserted to transfer data associated with a command into the asynchronous buffer. After the data has been moved into the buffer, an acknowledge signal may indicate that the transfer is complete. A launch signal may transfer the data in the asynchronous buffer to memory. Embodiments of the present invention allow the processing of a next command to begin at the earliest possible time while data associated with a previous command is being transferred into and out of the buffer, thereby increasing throughput and improving performance.

Optimizing data bandwidth across a variable asynchronous clock domain

A method and apparatus are provided for scaling an input bandwidth for bandwidth allocation technology. An original bandwidth count value of an input flow is received. A bandwidth scaler constant is provided and used for scaling the received original bandwidth count value to provide a scaled bandwidth value between zero and one. The scaled bandwidth value is stored and used for calculating a transmit probability for the input flow.

Method and apparatus for scaling input bandwidth for bandwidth allocation technology

A mechanism provided for controlling a transmission enable (TX_ENA) signal. The mechanism generates a queue of bits to track a sequence of commands and provides the transmit enable signal if the queue is empty. If an entry at a top of the queue indicates a write command, the mechanism provides the transmit enable signal for a predetermined number of cycles before the transmit enable signal is needed and until write data associated with the entry is transmitted, whereupon the entry is removed from the queue. If the entry at the top of the queue does not indicate a write command, the mechanism discontinues the transmit enable signal and removing the entry from the queue.

Handling of the transmit enable signal in a dynamic random access memory controller

A memory system employs calibrations to ensure the precise transmission of data. During calibrations, memory refreshes can occur; however, these refreshes can interfere with calibration streams. Therefore, to alleviate collisions and interferences, refreshes are deferred to periods where no calibrations are taking place. The number of deferred refreshes is also tracked such that the overall loss of refreshes is prevented.

Deferring refreshes during calibrations in memory systems

A structure of sequencers, a method, and a computer program are provided for performing initial and periodic calibrations in an XDR&#x2122; memory system. A memory controller that performs these calibrations is divided into identical, independent halves, with each half containing a Current/Impedance Calibration (i/z Cal) sequencer and six Bank sequencers. The i/z Cal sequencer contains three pathways that perform the XIO current and termination calibrations, and the XDR&#x2122; DRAM current and termination impedance calibrations. Each Bank sequencer contains normal read and write operation pathways that are reused to accomplish receive setup, receive hold, transmit setup, transmit hold, XIO receive, and XIO transmit timing calibrations. Initial and periodic calibrations are necessary to ensure the precise transfer of data between the XIOs and the XDR&#x2122; DRAMs.

Structure of sequencers that perform initial and periodic calibrations in a memory system

A memory controller design tool retrieves parameter ranges supported by a memory controller, and identifies troublesome parameter value combinations. The memory controller design tool suggests to 1) add logic to the memory controller to resolve the conflict, 2) incorporate a constraint that reduces/eliminates command collisions, data conflicts, and/or the need to check particular timing parameters, or 3) a combination of both. The memory controller design tool may work in conjunction with a memory controller designer to define and use the constraints.

Growing community of inventors

Rochester, MN, United States of America

Mark David Bellows