Parallel access virtual channel memory system

Details Parallel access virtual channel memory system Parallel access virtual channel memory system

1999-11-10

2002-11-05

Peikari, B. James (Department: 2186)

Electrical computers and digital processing systems: memory

Storage accessing and control

Hierarchical memories

C711S119000, C711S118000, C711S147000, C711S150000, C710S038000, C710S031000

Reexamination Certificate

active

06477621

ABSTRACT:

BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a memory system for a semiconductor memory. More particularly, the present invention relates to a memory system which provides a plurality of virtual access channels to facilitate access by a plurality of memory masters.
2. Description of the Prior Art
Conventional data processing systems typically include multiple processors/processes which share a system memory. The multiple processors/processes (i.e., memory masters) access the system memory (e.g., general system memory or graphic frame buffer memory) in a multi-tasking manner. The memory masters can include central processing units (CPUs), graphics processors, PCI bus masters and EISA/ISA bus masters. Each memory master accesses portions of the system memory which exhibit an address locality, a time locality and/or a particular block size. It would therefore be desirable to have a memory system which allows multiple memory masters to access a system memory in an efficient manner. It would further be desirable if such a memory system could be dynamically modified to accommodate different types of memory masters.
FIG. 1
is a block diagram of a multi-processing system
100
which employs a shared memory architecture. System
100
includes processors
101
a
-
101
c
, dedicated cache memories
102
a
-
102
c
, dedicated cache controllers
103
a
-
103
c
, system bus
104
, global main memory
105
and memory controller
106
. Processors
101
a
-
101
c
share main memory
105
through common parallel system bus
104
. Cache memories
102
a
-
102
c
are typically constructed using relatively high speed SRAM arrays. Main memory
105
is typically constructed using relatively low speed and low cost DRAM arrays. Systems such as system
100
are described in the following references: (1) “Protocols Keep Data Consistent”, John Gallant, EDN Mar. 14, 1991, pp.41-50 and (2) “High-Speed Memory Systems”, A. V. Pohm and O. P. Agrawal, Reston Publishing, 1983, pp.79-83.
Dedicated cache memories
102
a
-
102
c
reduce the frequency with which each of processors
101
a
-
10
c
access main memory
105
. This reduces the amount of traffic on system bus
104
. However, cache memories
102
a
-
102
c
are relatively expensive. In system
100
, an expensive cache memory must be added for each added processor. In addition, system
100
requires control logic to maintain the consistency of data in cache memories
102
a
-
102
c
and main memory
105
(i.e., cache coherence). The problem of cache coherence is described in more detail in “Scalable Shared Memory Multiprocessors”, M. Dubois and S. S. Thakkar, Kluwer Academic Publishers, 1992, pp.153-166. The control logic required to provide cache coherence increases the cost and decreases the performance of system
100
. In addition, the efficiency of main memory
105
and system bus
104
suffers if the data values fetched into cache memories
102
a
-
102
c
are not used.
FIG. 2
is a block diagram of another conventional multi-processor system
200
which includes a global main memory
204
which is divided into modules
206
a
-
206
c
. Each of main memory modules
206
a
-
206
c
is attached to a single corresponding cache memory module
205
a
-
205
c
, respectively. Each of cache memory modules
205
a
-
205
c
is attached to a main memory bus
202
. Processors
201
a
-
201
c
are also attached to main bus
202
. Processors
201
a
-
201
c
share cache memory modules
205
a
-
205
c
and main memory modules
206
a
-
206
c
. System
200
is described in, “High-Speed Memory Systems”, Pohm et al., pp.75-79. When the number of processors is approximately equal to the number of memory modules (i.e., cache memory modules), cache thrashing can occur. Cache thrashing refers to the constant replacement of cache lines. Cache thrashing substantially degrades system performance.
To minimize the cost of SRAM cache memories, some prior art systems use additional prefetch buffers for instructions and data. These prefetch buffers increase the cache-hit rate without requiring large cache memories. Such prefetch buffers are described in PCT Patent Application PCT/US93/01814 (WO 93/18459), entitled “Prefetching Into a Cache to Minimize Main Memory Access Time and Cache Size in a Computer System” by Karnamadakala Krishnamohan et al. The prefetch buffers are used in a traditional separate cache memory configuration, and memory bandwidth is consumed by both the prefetch operations and the caching operations. A robust prefetch algorithm (with a consistently high probability of prefetching the correct information) and an adequate cache size and organization (to provide a high cache hit rate) is required to deliver any performance improvement over traditional caching schemes.
Other conventional systems use the sense-amplifiers of a DRAM array as a cache memory. (See, e.g., PCT Patent Publication PCT/US91/02590, by M. Farmwald et al.) Using the sense-amplifiers of a DRAM array as cache memory provides low cost, high transfer bandwidth between the main memory and the cache memory. The cache hit access time, equal to the time required to perform a CAS (column access) operation, is relatively short. However, the cache miss access time of such a system is substantially longer than the normal memory access time of the DRAM array (without using the sense amplifiers as a cache memory). This is because when the sense amplifiers are used as cache memory, the DRAM array is kept in the page mode (or activated mode) even when the DRAM array is not being accessed. A cache miss therefore requires that the DRAM array perform a precharge operation followed by RAS (row access) and CAS (column access) operations. The time required to perform the precharge operation (i.e., the precharge time) is approximately twice as long as the time required to perform the RAS operation. The total memory access time is therefore equal to the sum of the precharge time, the RAS access time and the CAS access time of the DRAM array. In contrast, during normal operation of the DRAM array, the DRAM array is in precharged mode when it is not being accessed, and the memory access time is equal to the RAS access time plus the CAS access time of the DRAM array.
Another prior art cache memory system includes an SRAM cache memory which is integrated into a DRAM array. The DRAM array includes four banks which collectively serve as the main system memory. The SRAM cache memory includes a cache row register which has the capacity to store a complete row of data from one of the banks of the DRAM array. A last row read (LRR) address latch stores the address of the last row read from the DRAM array. When the row address of a current read access is equal to the row address stored in the LRR address latch, the requested data values are read from the row register, rather than the DRAM array. Thus, there is one cache entry in the cache row register which is shared by each of the four banks in the DRAM array. This prior art memory system is described in more detail in DM 2202 EDRAM 1 MB×4 Enhanced Dynamic RAM, Preliminary Datasheet, Ramtron International Corp., pp. 1-18.
It is therefore desirable to have a memory system which overcomes the previously described shortcomings of the prior art memory systems.
SUMMARY
In accordance with the present invention, a memory system includes a main memory and a plurality of virtual access channels connected in parallel to the main memory. The main memory typically includes a plurality of memory banks. Each of the virtual access channels includes a set of memory access resources for accessing the main memory. These memory access resources can include, for example, cache resources, burst access control resources, and memory precharge resources. Each of the virtual access channels is independently addressable by an external memory master.
By enabling the virtual access channels to be addressed by external memory masters, the virtual access channels can be flexibly assigned to serve different memory masters as required by the data processing system to which the

Affiliated with

Also associated with

Rating

Parallel access virtual channel memory system does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Parallel access virtual channel memory system, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Parallel access virtual channel memory system will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-2922881