Integrated processing and L2 DRAM cache

• US 5,895,487 A
• Filed: 11/13/1996
• Issued: 04/20/1999
• Est. Priority Date: 11/13/1996
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Matick, Richard Edward, Boyd, William Todd, Ignatowski, Michael, Schuster, Stanley Everett, Heller, Thomas James Jr.
• Primary Examiner(s): Chan, Eddie P.
• Assistant Examiner(s): Ellis, Kevin L.
• Application Number:
  US08/748,300
• Time in Patent Office: 888 Days
• Field of Search: 711/119, 711/121, 711/122, 711/124
• US Class Current: 711/122
• CPC Class Codes: G06F 12/0817 using directory methods
• CPC Class Codes: G06F 12/0851 Cache with interleaved addressing
• CPC Class Codes: G06F 12/0897 with two or more cache hierarchy levels with multilevel cache hierarchies G06F12/0811
• CPC Class Codes: G06F 12/1045 associated with a data cache
• CPC Class Codes: G06F 12/1054 the data cache being concurrently physically addressed
• CPC Class Codes: G06F 12/1063 the data cache being concurrently virtually addressed
• CPC Class Codes: G06F 15/7846 On-chip cache and off-chip main memory
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

An integrated processor and level two (L2) dynamic random access memory (DRAM) are fabricated on a single chip. As an extension of this basic structure, the invention also contemplates multiprocessor "node" chips in which multiple processors are integrated on a single chip with L2 cache. By integrating the processor and L2 DRAM cache on a single chip, high on-chip bandwidth, reduced latency and higher performance are achieved. A multiprocessor system can be realized in which a plurality of processors with integrated L2 DRAM cache are connected in a loosely coupled multiprocessor system. Alternatively, the single chip technology can be used to implement a plurality of processors integrated on a single chip with an L2 DRAM cache which may be either private or shared. This approach overcomes a number of issues which limit the performance and cost of a memory hierarchy. When the L2 DRAM cache is placed on the same chip as the processor, the time needed for two chip-to-chip crossings is eliminated. Since these crossings require off-chip drivers and receivers and must be synchronized with the system clock, the time involved is substantial. This means that with the integrated L2 DRAM cache, latency is reduced.

Claims

1. A single chip comprising:

multiple, independent processors;

each processor having a private L1 cache and associated translation/memory management logic to implement a set-associative, late-select cache, each L1 cache having a multi-ported cache directory for fast coherency maintenance via a fully shared Snoopy protocol;

the L1 cache directories being interconnected by a plurality of buses to allow simultaneous interrogation and updating and having a late-write capability for "same cycle" update;

outputs from all L1 caches being interconnected by a selector/cross-point-switch for transferring data between caches, each L1 cache having a pseudo-two-port structure with associated full-line width reload and store-back buffers, each cache input/output data width being equal to a full line and connected to an input/output bus of equal width for line transfer;

each processor having a private L2 cache with an interface to the private L1 cache of the processor;

each L2 cache having translation/management logic to implement a set-associative, late-select organization with DRAM directories having a late-write capability, each L2 cache comprising a DRAM main array, and an SRAM buffer to interface to the L1 cache;

coherency between the L2s being maintained by a global directory, selectors and logic directing cross-interrogates to an appropriate L2 cache, each L2 cache having a pseudo-two-port structure with data buffers for reload and store-back and an interface to main memory, a wide data input/output bus with a width equal to a cache line for reload/store back operations; and

logic and data circuits for interfacing to an external main memory management unit, said chip being capable of working alone as a single node system or coupled via an external controller to other identical or similar nodes.

2. A single chip comprising:

N independent processors;

each processor having a private L1 cache and associated translation/memory management logic to implement a set-associative, late-select cache, each L1 cache having a multi-ported cache directory for fast coherency maintenance via a fully shared Snoopy protocol;

the L1 cache directories being interconnected by a plurality of buses to allow simultaneous interrogation and updating and having a late-write capability for "same cycle" update;

outputs from all L1 caches being interconnected by a selector/cross-point-switch for transferring data between caches, each L1 cache having a pseudo-two-port structure with associated full-line width reload and store-back buffers, each cache input/output data width being equal to a full line and connected to an input/output bus of equal width for line transfer;

the N independent processors all sharing one on-chip L2 cache having a directory, with the L2 cache and directory both physically structured into N banks;

each bank of the L2 cache having translation/management logic to implement a set-associative, late-select organization, the L2 cache and directory having a late-write capability;

the L2 cache comprising a DRAM main array and a SRAM buffer to interface to the L1 cache;

coherency between the N L2 cache banks being maintained by a global directory with selectors and logic for directing cross-interrogates to the appropriate L2 bank;

a wide data input/output bus having a width equal to a cache line for reload/store back operations;

each bank of the L2 cache having an integrated address bus decoder-selector-priority function directly coupled to a data input/output selector-bus section; and

logic and data circuits for interfacing to an external main memory management unit, such chip being capable of working alone as a single node system, or coupled via an external controlled to other identical or similar nodes.

3. The single chip of claim 2 wherein each bank of the shared L2 cache independently accesses only its portion of the total L2 array;

each bank having a separate bank priority selector, bank address selector and data input/output selector;

each bank priority selector receiving a separate bank address from each of the N independent processors and, from these, generating one of N possible enable signals;

each bank address selector receiving a remaining portion of the memory addresses from each of the N independent processors, each bank address selector receiving N enable inputs, all from its respective bank priority selector, each bank priority selector asserting one or none of the enable output signals as an input for its bank address selector which provides the address to the bank array; and

each data input/output selector receiving N inputs, one from each array input/output port of the N banks, and N enable signals, one from each of the N bank priority selectors, the output of each data input/output selector serving as an input/output bus for only its respective processor.

4. The single chip of claim 2 wherein each bank of the shared L2 cache independently accesses only its portion of the total L2 array;

each bank having a separate bank priority selector, bank address selector and data input/output selector;

each bank priority selector receiving a separate bank address from each of the N independent processors and, from these, generating one of N possible enable signals;

each bank address selector receiving the remaining portion of the memory addresses from each of the N independent processors, each bank address selector receiving N enable inputs, all from its respective bank priority selector, each bank priority selector asserting one or none of the enable output signals as an input for its bank address selector which provides the address to the bank array;

each input/output selector having as data input only an input/output of its respective bank, there being N output ports, one for each of the N independent processors, a selection being determined by an assertion of one or none of the N enable inputs which all come from the same bank priority selector for each respective bank, the enable signal causing data from its respective array to be placed on one of N data output buses where each such bus is ORed to a corresponding bus of each of the other N-1 data input/output selectors, each one of the N data output buses serving one of the N independent processors.

5. A single chip comprising:

multiple, independent processors;

each processor sharing one large L1 cache and associated translation/memory management unit with logic to implement a set-associative, late-select cache, the L1 cache having a directory, and L1 cache array and directory being multiported with as many ports as there are processors on-chip, each port having an independent address and data buses, one for each processor to allow simultaneous access and updating of the L1 cache array and directory, with both having a late-write capability for "same cycle" update;

L1 cache input/output data width equal to a full line size and connected to an equal width bus for line transfer;

the multiple, independent processors all sharing one on-chip L2 cache having a directory, with the L2 cache and directory both physically structured into as many banks as there are processors on-chip;

each bank of the L2 cache having translation/management logic to implement a set-associative, late-select organization, the L2 cache and directory having a late-write capability;

each L2 cache comprising a DRAM main array and a SRAM buffer to interface to the L1 cache;

coherency between the L2 cache banks being maintained by a global directory with selectors and logic for directing cross-interrogates to the appropriate L2 bank;

a wide data input/output bus having a width equal to a cache line for reload/store back operations;

each bank of the L2 cache having an integrated address bus decoder-selector-priority function directly coupled to a data input/output selector-bus section; and

logic and data circuits for interfacing to an external main memory management unit, such chip being capable of working alone as a single node system, or coupled via an external controlled to other identical or similar nodes.

6. A multiprocessor system comprising:

a plurality of node chips, each containing multiple processors and L1 and L2 caches on a single chip, wherein said L1 cache is a SRAM and said L2 cache is a DRAM; and

an external controller chip connecting said plurality of node chips to form a larger multiprocessor system, an amount of multiprocessing being scalable by adding more or less node chips to said controller chip,

wherein at least one of said plurality of node chips further includes;

in each of said multiple processors a private L1 cache and associated translation/memory management logic to implement a set-associative, late-select cache, each private L1 cache having a multi-ported cache directory for fast coherency maintenance via a fully shared Snoopy protocol;

the L1 cache directories being interconnected by a plurality of buses to allow simultaneous interrogation and updating and having a late-write capability for "same cycle" update;

outputs from all private L1 caches of said at least one of said plurality of node chips being interconnected by a selector/cross-point-switch for transferring data between caches, each private L1 cache having a pseudo-two-port structure with associated full-line width reload and store-back buffers, each cache input/output data width being equal to a full line and connected to an input/output bus of equal width for line transfer;

each processor having a private L2 cache with an interface to the private L1 cache of the processor, each private L2 cache having translation/management logic to implement a set-associative, late-select organization with DRAM directories having a late-write capability, each private L2 cache comprising a DRAM main array, and an SRAM buffer to interface to the private L1 cache;

coherency between the private L2s being maintained by a global directory, selectors and logic directing corss-interrogates to an appropriate private L2 cache, each private L2 cache having a pseudo-two-port structure with data buffers for reload and store-back and an interface to main memory, a wide data input/output bus with a width equal to a cache line for reload/store back operations; and

logic and data circuits for interfacing to an external main memory management unit, said at least one of said plurality of chips being capable of working alone as a single node system or coupled via said external controller to other ones of said plurality of node chips.

7. The multiprocessor system recited in claim 6 wherein the L2 cache includes a SRAM array which acts as a high-speed front end buffer to the DRAM.

8. The multiprocessor system of claim 7 wherein said L1 and L2 caches are each composed of single, shared caches for the multiple processors on the chip.

9. The multiprocessor system of claim 7 wherein said L1 cache is composed of a plurality of private L1 caches, one for each of said multiple processors.

10. The multiprocessor system of claim 9 wherein said L2 cache is composed of a plurality of private L2 caches, one for each of said multiple processors, further comprising an on-chip global cross-interrogate directory for said private L2 caches.

11. The multiprocessor system of claim 9 wherein said L2 cache is composed of a single, shared L2 cache for the multiple processors on the chip.