Electrical computers and digital processing systems: memory – Storage accessing and control – Specific memory composition
Reexamination Certificate
1997-11-14
2001-05-01
Bragdon, Reginald G. (Department: 2186)
Electrical computers and digital processing systems: memory
Storage accessing and control
Specific memory composition
C711S164000, C707S793000
Reexamination Certificate
active
06226710
ABSTRACT:
BACKGROUND OF THE INVENTION
This invention relates generally to computer memory storage devices and corresponding controllers, and more particularly to a controller for associative memory or content addressable memory devices in which the contents of well-known random access memory (“RAM”) devices are retrieved not by use of location addresses, but by use of “keys” associated with the contents of corresponding memory locations.
A computer typically comprises a signal processor, which operates according to defined instructions. The computer also includes a memory storage device that stores the instructions for the signal processor along with data utilized by the processor for various purposes. The most common type of memory device is RAM, which stores data at particular locations normally defined by specific addresses. To either store or retrieve data in RAM, the signal processor must supply the specific address of the desired memory location to a memory address register associated with RAM. All memory locations within RAM are truly randomly-accessible in that the processor can access any location independently of all other locations in a specific period of time that is constant for all memory addresses.
A RAM device typically comprises a matrix of memory locations arranged by rows and columns. As such, these locations are indexed by row and column numbers. This arrangement of memory locations allows the RAM device to be of relatively high density and low cost. Modem RAM devices are also relatively fast in implementing data storage and retrieval operations. These factors contribute greatly to the large popularity of RAM devices.
With RAM, the number of discrete, addressable locations is essentially limited by the number of address lines provided with the memory device. For example, sixteen individual address lines allow for 2
16
, or 65536, separate and distinct memory locations to be addressed within a RAM integrated circuit. Modem commercially-available RAM devices may have at least one million (i.e., “1 Meg”) or more of individually-addressable memory locations.
On the other hand, generally the bit size or width of RAM locations varies greatly between commercially-available devices. The width of each memory location is typically determined, in part, by the width (i.e., number of bits) of the data bus employed by the computer system. For example, for a 16-bit wide data bus, the width of a RAM memory location may be 16 bits or may be 64 bits (i.e., four times the 16-bit word size). In the alternative, the word size may be only one bit wide. These sizes are purely exemplary. The word size depends primarily on the chosen application for the RAM device.
RAM devices are generally divided into two categories according to their electrical operating characteristics: static and dynamic. A static RAM (“SRAM”) device is constructed internally such that once data have been written into the locations, the logic states of the data are maintained as long as electrical power is applied to the SRAM device. On the other hand, a dynamic RAM (“DRAM”) device requires constant refreshing of its internal circuits to maintain the logic states of the stored data in the memory cells over time.
Despite their popularity, RAM devices have inherent limitations. The primary limitation involves accessing stored data using a location-address method. For example, if it is desired to access one specific piece of data, then, for fastest data retrieval, the corresponding RAM location must be known. This requirement stems from the fact that the RAM location address has no logical relationship to the stored data. In RAM, the address is merely an artificial construct. If the specific RAM address is unknown (which is often the case), the processor must employ some type of search process to locate the desired piece of data. With RAM, such searching is typically carried out sequentially, one location at a time.
This sequential searching drawback is magnified in certain software applications that store a large number of items in a data structure such as a table. Many modern software applications are table driven. Software developers have embraced this approach because the resulting software is flexible, understandable and readily maintainable. Generally, these applications consume considerable processor resources. For example, simulation applications are generally table intensive, with a reputation for consuming large amounts of processor time. Unfortunately, this means that the signal processor is spending a great deal of time just finding data entries in the tables.
More specifically, when it is desired to search through, correlate and/or sort a table of data, the signal processor must sequentially scan large blocks of RAM locations to find and/or position certain data during these operations. This serial data processing function is required since a RAM device generally has no capability for scanning, correlating and/or sorting its entire contents in parallel.
For example, in a data table implemented in RAM and containing a list of records (e.g., a list of people), with each record having several data fields (e.g., names, addresses, phone numbers), the signal processor must specify the exact RAM address to find the desired data in the table. Alternatively, the processor could run software that sequentially searches the entire table for a desired field. As compared to a location address, the concept of a “field” has somewhat more of a logical relationship to the stored data. Yet, because RAM generally lacks a parallel scanning ability built into the hardware logic on the RAM device, the signal processor must sequentially search the entire table to locate the desired data. Sequential searching is extremely time-consuming when utilizing RAM devices to store databases. In some simple data capture applications, the tedious implementation of software lookup algorithms represents a significant portion of the application's complexity. The inherent RAM “bottleneck” problem often means that the speed of the RAM device, in storing and accessing data, is the limiting speed factor in the overall computer system. This is becoming more evident as recent speed improvements in signal processors have advanced past speed improvements for memory devices.
As a way of solving the inherent performance penalty associated with the sequential processing of data tables stored in RAM devices, it has long been desired to make memory devices more intuitive in terms of storing and accessing data. That is, it is desired to provide a memory device that functions in more of an associative manner. This is akin to human memory, where stored abstract information is referenced not by addresses, but by some other logically-related abstract information. Ideally, the associative memory should also compare to RAM in terms of speed, cost and density.
Various solutions have been proposed to make computer memory devices more associative in nature while retaining speed, cost and density advantages. These solutions center around both software and hardware techniques. Software schemes for making RAM more associative typically involve such techniques as hashing algorithms, software data structures, databases and neural networks. While these techniques have had some success, they all have a speed cost associated with them, because each associative reference requires many RAM access cycles and, correspondingly, many signal processor cycles. Nevertheless, when utilizing these software techniques, signal processor and memory speed improvements have generally kept pace with application speed requirements.
For example, “hashing” generally refers to software algorithms basically used to store and retrieve data from memory devices. Generally, hashing algorithms randomly scatter data throughout the available memory space using various mathematical functions, such as simple multiply or divide operations. Essentially, hashing is the opposite of the orderly sorting of data in sequential memory locations. The same mathematical function is used to retrieve the stored data.
Bragdon Reginald G.
Morgan & Finnegan L.L.P.
Tran Denise
UTMC Microelectronic Systems Inc.
LandOfFree
Content addressable memory (CAM) engine does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Content addressable memory (CAM) engine, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Content addressable memory (CAM) engine will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2453411