International Business Machines Corporation

John E Barth

Howard L Kalter

Erik L Hedberg

Jeffrey H Dreibelbis

Claude Louis Bertin

William R Tonti

Wayne Frederick Ellis

Wayne John Howell

Donald Lawrence Wheater

Dharmendra S Modha

Kirk David Peterson

Toshiaki Kirihata

Philip George Emma

Subramanian Srikanteswara Iyer

Herbert Lei Ho

Babar Ali Khan

John Vernon Arthur

Vijayalakshmi Srinivasan

Hillery C Hunter

Andrew S Cassidy

Paul A Merolla

Jun Sawada

Matthew R Wordeman

Peter Anthony Sandon

Rodrigo Alvarez-Icaza Rivera

Bryan L Jackson

Andrew A Turner

John B DeForge

Arnold S Tran

Jennifer E Appleyard

Embodiments of the invention relate to processor arrays, and in particular, a processor array with interconnect circuits for bonding semiconductor dies. One embodiment comprises multiple semiconductor dies and at least one interconnect circuit for exchanging signals between the dies. Each die comprises at least one processor core circuit. Each interconnect circuit corresponds to a die of the processor array. Each interconnect circuit comprises one or more attachment pads for interconnecting a corresponding die with another die, and at least one multiplexor structure configured for exchanging bus signals in a reversed order.

Interconnect circuits at three-dimensional (3-D) bonding interfaces of a processor array

A semiconductor structure is disclosed in which, in an embodiment, a first substrate includes at least one buried plate disposed in an upper part of the first substrate. Each of the at least one buried plate includes at least one buried plate contact, and a plurality of deep trench capacitors disposed about the at least one buried plate contact. A first oxide layer is disposed over the first substrate. The deep trench capacitors and buried plate contacts in the first substrate may be accessed for use in a variety of memory and decoupling applications.

Creating deep trenches on underlying substrate

A hardware description language (HDL) design structure embodied on a machine-readable data storage medium includes elements that when processed in a computer aided design system generates a machine executable representation of a device for implementing dynamic refresh protocols for DRAM based cache. The HDL design structure further includes a DRAM cache partitioned into a refreshable portion and a non-refreshable portion; and a cache controller configured to assign incoming individual cache lines to one of the refreshable portion and the non-refreshable portion of the cache based on a usage history of the cache lines; wherein cache lines corresponding to data having a usage history below a defined frequency are assigned by the controller to the refreshable portion of the cache, and cache lines corresponding to data having a usage history at or above the defined frequency are assigned to the non-refreshable portion of the cache.

Structure for implementing dynamic refresh protocols for DRAM based cache

A memory architecture that utilizes single-ended dual-port destructive write memory cells and a local write-back buffer is described. Each cell has separate read and write ports that make it possible to read-out data from cells on one wordline in the array, and subsequently write-back to those cells while simultaneously reading-out the cell on another wordline in the array. By implementing an array of sense amplifiers such that one amplifier is coupled to each read bitline, and a latch receiving the result of the sensed data and delivering this data to the write data lines, it is possible to 'pipeline' the read-out and write-back phases of the read cycle. This allows for a write-back phase from one cycle to occur simultaneously with the read-out phase of another cycle. By extending the operation of the latch to accept data either from the sense amplifier, or from the memory data inputs, modified by the column address and masking bits, it is also possible to pipeline the read-out and the modify-write-back phases of a write cycle, allowing them to occur simultaneously. The architecture preferably employs a nondestructive read memory cell such as 2T or 3T gain cells, achieving an SRAM-like cycle and access times with a smaller and more SER immune memory cell.

High performance gain cell architecture

Wafer test and burn-in is accomplished with state machine or programmable test engines located on the wafer being tested. Each test engine requires less than 10 connections and each test engine can be connected to a plurality of chips, such as a row or a column of chips on the wafer. Thus, the number of pads of the wafer that must be connected for test is substantially reduced while a large degree of parallel testing is still provided. The test engines also permit on-wafer allocation of redundancy in parallel so that failing chips can be repaired after burn-in is complete. In addition, the programmable test engines can have their code altered so test programs can be modified to account for new information after the wafer has been fabricated. The test engines are used during burn-in to provide high frequency write signals to DRAM arrays that provide a higher effective voltage to the arrays, lowering the time required for burn-in. Connections to the wafer and between test engines and chips are provided along a membrane attached to the wafer. Membrane connectors can be formed or opened after the membrane is connected to the wafer so shorted chips can be disconnected. Preferably the membrane remains on the wafer after test, burn-in and dicing to provide a chip scale package. Thus, the very high cost of TCE matched materials, such as glass ceramic contactors, for wafer burn-in is avoided while providing benefit beyond test and burn-in for packaging.

Structures for wafer level test and burn-in

An embedded memory macro device includes a memory system and a logic circuit constructed on a common semiconductor substrate. The memory system and logic circuit communicate through a handshake procedure via a system data interlock signal. During read cycles the memory system uses the system data interlock signal to tell the logic circuit when data at memory system data outputs is valid. In the preferred embodiment, during write cycles the memory system uses the system data interlock signal to tell the logic circuit when data has been successfully written. The logic circuit needs to wait only until the system data interlock signal permits it to proceed. It then signals the memory system to reset the system data interlock signal and can immediately initiate a new read or write cycle.

Interactive method for self-adjusted access on embedded DRAM memory

A method is provided for flash writing to multiple cells of a memory array. Initially, a first set of word lines, each of which controls connection of a memory cell of a first set of memory cells to a first bit line of a bit line pair, is turned on. The voltage between the two bit lines of the bit line pair is then equalized so that the charge on the first bit line of the bit line pair is higher than the charge on the second bit line of the bit line pair. Next, a sense amplifier attached to the bit line pair is turned on to sense a difference in charge between the bit line pair and to charge the first set of memory cells. Then a second set of word lines, each of which controls connection of a memory cell of a second set of memory cells to the second bit line is turned on. Finally, the word lines previously turned on are shut off and then the sense amplifier is shut off. Additionally, an apparatus is provided which allows for: turning on multiple word lines at one time, keeping an equalization means on until after at least one set of word lines have gone on, and controlling the operation of the sense amplifier to turn on and shut off at appropriate times.

Method for setting test voltages in a flash write mode

Growing community of inventors

Williston, VT, United States of America

John E Barth