Dokumentenidentifikation EP0797148 24.10.2002
EP-Veröffentlichungsnummer 0797148
Titel Cache-Struktur mit mehreren Modi
Anmelder Sharp K.K., Osaka, JP;
Sharp Microeletronics Technology, Inc., Camas, Wash., US
Erfinder Spaderna, Dieter, Portland, Oregon 97229, US
Vertreter Müller - Hoffmann & Partner Patentanwälte, 81667 München
DE-Aktenzeichen 69715491
Vertragsstaaten DE, FR, GB
Sprache des Dokument EN
EP-Anmeldetag 03.03.1997
EP-Aktenzeichen 973013808
EP-Offenlegungsdatum 24.09.1997
EP date of grant 18.09.2002
Veröffentlichungstag im Patentblatt 24.10.2002
IPC-Hauptklasse G06F 12/08

Field of the Invention

The invention relates to integrated circuits, and especially to an integrated circuit which includes a reduced instruction set computer (RISC) central processing unit (CPU) which is intended for use in portable devices which incorporates a block of on-chip memory which may be used as a cache or as RAM.

Background of the Invention

Portable devices include electronic instruments such as personal information manager, cellular telephones, digital cameras, hand-held games, bar-code scanners, medical equipment, electronic instrumentation, and navigation systems, specifically global positioning satellite navigation systems.

To be commercially successful, portable devices require integrated circuits which are low in cost, have low power requirements to insure long battery life, and have high standards of performance to insure that their output is accurate and usable. Additionally, the interface between the integrated circuit and the remainder of the portable device must be of a plug-and-play design so that a single type of integrated circuit may be used with a number of portable devices. Part of such flexibility is that the designer of a portable device must be able to easily incorporate the IC into the device as an off-the-shelf component, which does not require any internal modification to be used in a variety of applications.

The "computing power" found in integrated circuit CPUs, such as the Intel 30*86 and Pentium® series chips, and the Motorola 68000 series chips, has increased significantly over the last few years. At the same time, the size of such chips has grown significantly, as has their power requirements. Such chips are designated as conventional instruction set computers (CISC) and have come to require significant blocks of associated random access memory (RAM) while the applications that have been written to run on computers containing these chips have grown, seemingly without limit, requiring vast amounts of hard drive space. Such CISC devices are not easily usable with portable devices due to their power requirements and size.

Reduced instruction set computers (RISC) were originally used in high-end graphics applications, and in CAE/CAD work stations. The RISC architecture, however, enables an IC to have a significantly smaller die size, because the smaller instruction set of the RISC technology requires fewer transistors to implement, which leads to simpler designs which consequently take less time to complete and to debug. Additionally, smaller chips having shorter signal paths mean that each instruction cycle is of a shorter duration. The relative size of a RISC CPU is significantly smaller than a CISC CPU, for instance, the Intel 386 SL chip is approximately 170 mm2, while a RISC chip having similar computational abilities is slightly more than 5 mm2.

The small size of RISC-based CPUs makes the RISC architecture ideal for "system-on-chip" (SOC) applications, wherein the CPU and a number of other structures are located on a single chip. Such a SOC architecture may result in a chip that is still considerably smaller than a CISC CPU, but which contains all of the computational and control structures on a single integrated circuit. A SOC architecture will generally include the RISC CPU and some type of local RAM and/or data cache. Additionally, the chip may include internal and external bus controllers, various types of communication ports, an interrupt controller, and pulse width modulator, various configuration registers, various timer/counter structures, and some type of output controller, such as an LCD controller. Such a structure may be configured in a 32-bit architecture, with associated peripherals integrated onto the chip, which integration allows the designer of the portable device incorporating the chip to reduce the development cycle and accelerate the product introduction to market. The chip structure may have an external 16-bit data bus with an integrated, programmable bus controller capable of supporting 8 or 16-bit SRAM, DRAM, EPROM and/or memory devices, which do not require additional buffers in order to function with the integrated circuit. The chip may be operated at either 3.3 volts or 5 volts, which will require between 100 mW and 350 mW, respectively.

By combining a number of peripherals on the chip, and providing an internal bus amongst the CPU and peripherals, it is possible to conduct a number of operations on chip, while simultaneously controlling off-chip operations, such as memory stores and retrieves.

The integrated circuit may include a memory interface which provides multiple programmable chip enables, allowing users to set wait states and memory width, 8 or 16-bits wide. The integrated circuit provides for address decoding and DRAM control logic, which allows an external bus master to perform data transfers without requiring external address decoding or external DRAM controllers. In the case where sequential accesses are being used, the integrated circuit automatically increments the initial memory address supplied by the external bus master, thereby speeding transfers.

US-A-5 410 669 discloses a data processing system having a dual purpose memory comprising multiple cache sets. Each cache can be individually configured as either a cache set or an SRAM bank.

EP-A-0 529 217 discloses an integrated circuit chip having a memory section and switch means for switching the memory between a first standard mode of operation and a second cache mode, where a user can switch back and forth between the two operating modes, preserving data or flushing it when needed. This may enable the user to such different tasks in different operating modes.

The specific subject of the instant application is a multimode cache structure which includes a predefined block of memory and controls for that block of memory which allow the memory block to perform multiple functions. The selectable, multiple functions include a cache mode, a SRAM. mode, a flush mode and an invalidate mode. A control register is defined and is associated with the predefined memory block, which control register includes multiple status bits therein. Each of the status bits corresponds to one of the multiple functions and, when a particular status bit is set, the predefined block of memory performs a function corresponding to the status bit that is set.

It is desirable to provide an integrated circuit having a CPU wherein a block of memory is capable of performing more than one function.

It is also desirable to provide a memory block that may act as a cache or as a SRAM.

It is also desirable to provide an integrated circuit having a CPU which allows a user to select the functions of memory blocks on the integrated circuit.

The invention provides a multimode cache structure as set out in claim 1.

Advantages of the invention will become more fully apparent as the description which follows is read in connection with the drawings.

Brief Description of the Drawings

  • Fig. 1 is a block diagram of the system on chip structure of an embodiment of the invention.
  • Fig. 2 is a block diagram of a memory block in a cache mode.
  • Fig. 3 is a representation of a control register of an embodiment of the invention.
  • Fig. 4 is a representation of a segment descriptor register of an embodiment of the invention.

Best Mode of Practicing the Invention

Referring initially to Fig. 1, a system-on-chip (SOC) structure of an embodiment of the invention is depicted generally at 10. IC 10 includes a RISC CPU 12, which may include an embedded microcontroller 12a therefor, which is connected to a 32-bit internal bus 14. CPU 12 is directly connected to internal bus controller 16, which in turn is connected to bus 14 and to a bus map 18. IC 10 further includes a local RAM (SRAM) 20, a combined instruction/data cache 22, also referred to herein as a predefined block of memory, and an external access port 24. In the preferred embodiment, IC 10 is intended to connect to a liquid crystal display (LCD) and to that end, includes a LCD controller 26 which is connected to an LCD panel interface 28. An external bus controller 30 is provided and is connected to an external memory interface 32 and external chip selects 34, which, in the preferred embodiment, are part of external bus controller 30. Memory interface 32 and external chip selects 34 are connected by various buses to external memory, which may include SRAM and DRAM portions (not shown) and to various peripheral devices, such as I/O devices, hard drives, etc.

A number of configuration registers 36 are provided, which may be located in internal bus controller 16. A number of on-chip internal "peripherals" are connected to internal bus 14, and include a universal asynchronous receiver/transmitter (UART) 38, a parallel port 40, a timer/counter 42, an interrupt controller 44, and a pulse width modulator (PWM) 46.

Referring now to Fig. 2, a combined instruction/data cache is depicted in greater detail. As previously noted, the block of memory which makes up cache 22 is divided into two portions; the first portion is a Tag RAM portion 48, and the second portion is a data RAM portion 50. Associated with memory block 22 is a cache control register 54 which, in the preferred embodiment, is located in internal bus controller 16.

Cache 22, in the preferred embodiment, is a 2K-byte cache. The cache is 4-way set associative, with 128 sets, as indicated in Fig. 2 by sets &sl0;-127. In the preferred embodiment, each set includes four 1-word lines of cached data, with each line capable of storing a 32-bit word in data RAM portion 50 of memory block 22. Each set also includes four lines of 3-bit status 48a, and 23-bit address 48b in the Tag RAM portion of memory block 22.

Status bits 48a include an M-bit which indicates whether or not the data in the cache line has been modified or not, which is set to &sl0; when data and the associated address is written into the line, and which is set to 1 if the data is modified. A second status bit is a V-bit, which indicates whether the data is valid or not. If the V-bit is set to 1, the indication is that the data associated therewith is valid, if the V-bit is set to &sl0;, the data is considered to be invalid. The third bit is a least recently used (LRU) bit, and is used to trace the history of the sets in the cache.

Address register 52 is a 32-bit register. Bits 1 and &sl0; are not used. Bits 2 through 8 are used to identify the set number (&sl0;-127) while bits 9 through 31 provide the address in Tag RAM 48a.

In normal operation. an address is transferred from RISC CPU 12 to address register 52. The four lines in Tag RAM 48 are interrogated in parallel. If any of the four lines match the address tag in address register 52, the corresponding data from data RAM 50 is output. Memory block 22, when functioning as a cache, uses a write-back protocol. When a cache write occurs, the entry is marked as modified, by the appropriate status bit. When that entry is later replaced, the main memory (off chip) is updated. The write-back protocol reduces the traffic over internal bus 14 and through external bus controller 30, however, the main memory is left with old data until it is updated by the cache.

Referring momentarily to Fig. 3, cache control register (CCR) 54, which is located in internal bus controller 16, in the preferred embodiment, is an 8-bit register which controls the functionality of memory block 22. Only four bits of the register are used, and those four bits are designated as E, S, F, and I. When the E-bit is set to logical 1, the cache mode of memcry block 22 is enabled. When the S-bit of register 54 is set to logical 1, the SRAM mode is enabled. When the F-bit of register 54 is set to 1, the flush mode is enabled. When the I-bit of register 54 is set to logical 1, the cache is declared to be invalid. Bits E, F and S of register 54 are mutually exclusive, i.e., only one of them may be set to logical 1 at any time. Memory block 22 is placed in cache mode following a reset operation.

Referring back to Fig. 2, à segment descriptor register (SDR) is depicted at 56. SDR 56 is used when bit S in CCR 54 is set to logical 1, and memory block 22 is in its SRAM, unified storage region, mode. Again, bits &sl0; and 1 of SDR 56 are not used. Bits 2-8 determine which set is to be addressed, and bits 9 and 1&sl0; determine which line is to be addressed. Bits 11-31 provide the address locations in memory block 22 which are to be addressed. When memory block 22 is in the SRAM mode, Tag RAM 48 and Data RAM 50 are accessed as a 1K-word SRAM. This feature allows a system designer to use memory block 22 as an extra on-chip SRAM, and also allows the designer to examine the cached data, the address tags and the status bits. When memory block 22 is in SRAM mode, the data RAM, which is 512 words long in the preferred embodiment, will map to the initial address locations, and the Tag RAM, which is also 512 words long, will map to the sequential address locations. Although each line of the Tag RAM is only 26 bits wide, the Tag RAM lines will be accessed as if they were a 32-bit word, with six zeros appended to the left (right justified) of the 3-bit status and 23-bit address tag. All accesses to memory block 22 in SRAM mode must be of word length, as byte accesses are not allowed in the preferred embodiment.

The other modes, i.e., flush mode and invalidate mode, are used to manage memory block 22. When block 22 is in flush mode, any access to a cache set wherein any of the four lines thereof have been modified, will result in the modified line(s) being written back to main memory. Therefore, the contents of the cache may be forced into main memory by accessing all 128 sets in the cache sequentially. This feature is particularly well-suited for direct memory access (DMA) protocols on a chacheable memory segment.

If the cache is in invalidate mode, all of the lines in the cache are invalidated. Any memory access by CPU 12 will be forced to go to main memory (off-chip).

Thus, a memory block for use with a RISC CPU has been disclosed, wherein the memory block may be used as a cache or as a SRAM. In the case when the memory block is used as a cache, it provides a zero-wait state on the chip whenever a cache hit occurs, and minimizes delays on cache misses. Memory updates are done in a write-back protocol to reduce traffic on the external bus. The cache uses an LRU algorithm for any replacement protocoL When the memory block is in a SRAM mode, it provides extra on-chip RAM for use in suitable applications. Such applications may make use of memory block 22 for RAM functions when the application which is being run does not require rapid access to cached data. Such applications may be used in portable devices such as digital cameras and peripheral controllers in which on-chip SRAM is more important than the caching of programming instructions.

  1. Cache-Struktur mit mehreren Modi für eine integrierte Schaltung mit einem Speicherblock (22), bei der zwischen mehreren Modi, einschließlich eines Cachemodus, in dem eine Anzahl von Cachegruppen mit jeweils mehreren Cachezeilen im Speicherblock abgespeichert wird, eines SRAM-Modus und eines Ungültigmachmodus, in dem alle Cachezeilen ungültig gemacht werden und alle Speicherzugriffe auf den Hauptspeicher gelenkt werden, umgeschaltet werden kann, und einem Steuerregister (54) mit einer Anzahl von Statusbits, das dem Speicherblock (22) zugeordnet ist, dadurch gekennzeichnet, dass zu den mehreren Modi auch ein Ausräummodus gehört, bei dem ein Zugriff auf eine Cachegruppe, die seit einem vorigen Zugriff modifiziert wurde, dazu führt, dass die modifizierte Cachegruppe in den Hauptspeicher eingeschrieben wird, wobei der Cache-, der SRAM- und der Ausräummodus einander wechselseitig ausschließen, und dass die mehreren Statusbits im Steuerregister dem Cache-, dem SRAM-, dem Ausräum- und dem Ungültigmachmodus entsprechen, wobei, wenn ein Statusbit gesetzt ist, der Speicherblock (22) den dem gesetzten Statusbit entsprechenden Modus ausführt.
  2. Cache nach Anspruch 1, bei dem im SRAM-Modus sowohl ein Tag-RAM-Teil als auch ein Daten-RAM-Teil auf Adressorte abgebildet sind.
  3. Cache nach Anspruch 1 oder Anspruch 2, bei dem der Speicherblock (22) im Cache-Modus über einen Tag-RAM-Teil (48) und einen Daten-RAM-Teil (50) verfügt und er im SRAM-Modus über einen vereinheitlichten Speicherbereich verfügt.
  4. Cache nach Anspruch 3, bei dem das Steuerregister (54) ein Segmentdeskriptorregister (56) ist, das verwendet wird, wenn sich der Speicherblock (22) im SRAM-Modus befindet.
  5. Cache nach einem der vorstehenden Ansprüche, bei dem sich der Speicherblock (22) auf einen Rücksetzvorgang hin im Cachemodus befindet.
  6. Cache nach einem der vorstehenden Ansprüche, bei dem Zugriffe mit der Einheit einer Wortlänge erfolgen, wenn sich der Speicherblock (22) im SRAM-Modus befindet.
  7. Integrierte Schaltung mit einer Cache-Struktur mit mehreren Modi nach einem der vorstehenden Ansprüche.
  8. Integrierte Schaltung nach Anspruch 7 mit einer CPU und einer Verbindung (32) zu einer Hauptspeichereinheit.
  1. A multimode cache structure for an integrated circuit comprising a block of memory (22) and which can be shifted between a multiplicity of modes including a cache mode, wherein a plurality of cache sets each having a plurality of cache lines are stored in said memory block, an SRAM mode and an invalidate mode wherein in invalidate mode, all cache lines are invalidated and all memory access is directed to main memory, and a control register (54) including a multiplicity of status bits associated with said memory block (22), characterised in that the multiplicity of modes also includes a flush mode, wherein in flush mode, access to a cache set that has been modified since a previous access causes the modified cache set to be written to main memory, said cache, SRAM and flush modes being mutually exclusive, and in that said multiplicity of status bits in said control register (54), correspond to said cache, SRAM, flush and invalidate modes, wherein, with a status bit set, said memory block (22) performs in a mode corresponding to the set status bit.
  2. The cache of claim 1 wherein in said SRAM mode, both a Tag RAM portion and a data RAM portion are mapped to address locations.
  3. The cache of claim 1 or claim 2 wherein the memory block (22) includes a Tag RAM portion (48) and a data RAM portion (50) in cache mode, and wherein said memory block (22) includes a unified storage region in SRAM mode.
  4. The cache of claim 3 wherein said control register (54) is a segment descriptor register (56) used when said memory block (22) is in SRAM mode.
  5. The cache of any preceding claim wherein said memory block (22) is in cache mode following a reset operation.
  6. The cache of any preceding claim wherein accesses are in word-length units when said memory block (22) is in SRAM mode.
  7. An integrated circuit comprising a multimode cache structure as claimed in any preceding claim.
  8. The integrated circuit of claim 7 having a CPU and a connection (32) to a main memory unit.
  1. Structure de mémoire cache multimode pour un circuit intégré comprenant un bloc de mémoire (22) et qui peut être décalée entre une multiplicité de modes comprenant un mode de mémoire cache, dans laquelle plusieurs ensembles de mémoire cache ayant chacun plusieurs lignes de mémoire cache sont stockés dans ledit bloc de mémoire, un mode de SRAM et un mode d'invalidation dans lesquels, dans le mode d'invalidation, toutes les lignes de mémoire cache sont invalidées et tous les accès de mémoire sont dirigés vers la mémoire principale, et un registre de commande (54), comprenant une multiplicité de bits d'état, associé audit bloc de mémoire (22), caractérisée en ce que la multiplicité de modes comprend également un mode de vidage, dans lequel l'accès à un ensemble de mémoire. cache qui a été modifié depuis un accès précédent provoque l'écriture de l'ensemble de mémoire cache modifié dans la mémoire principale, lesdits modes de mémoire cache, de SRAM et de vidage étant mutuellement exclusifs, et en ce que ladite multiplicité de bits d'état dans ledit registre de commande (54) correspond auxdits modes de mémoire cache, de SRAM, de vidage et d'invalidation, dans lesquels, un bit d'état étant mis à un, ledit bloc de mémoire (22) fonctionne dans un mode correspondant au bit d'état mis à un.
  2. Mémoire cache selon la revendication 1, dans laquelle, dans ledit mode de SRAM, à la fois une partie RAM d'Etiquette et une partie RAM de données sont mises en correspondance avec des emplacements d'adresse.
  3. Mémoire cache selon la revendication 1 ou la revendication 2, dans laquelle le bloc de mémoire (22) comprend une partie RAM d'Etiquette (48) et une partie RAM de données (50) dans le mode de mémoire cache, et dans laquelle ledit bloc de mémoire (22) comprend une zone de stockage unifiée dans le mode de SRAM.
  4. Mémoire cache selon la revendication 3, dans laquelle ledit registre de commande (54) est un registre de descripteur de segment (56) utilisé quand ledit bloc de mémoire (22) est dans le mode de SRAM.
  5. Mémoire cache selon l'une quelconque des revendications précédentes, dans laquelle ledit bloc de mémoire (22) est dans un mode de mémoire cache après une opération de remise à l'état initial.
  6. Mémoire cache selon l'une quelconque des revendications précédentes, dans laquelle les accès sont en unités de longueur de mot quand ledit bloc de mémoire (22) est dans le mode de SRAM.
  7. Circuit intégré comprenant une structure de mémoire cache multimode selon l'une quelconque des revendications précédentes.
  8. Circuit intégré selon la revendication 7 ayant une CPU et une connexion (32) à une unité de mémoire principale.

A Täglicher Lebensbedarf
B Arbeitsverfahren; Transportieren
C Chemie; Hüttenwesen
D Textilien; Papier
E Bauwesen; Erdbohren; Bergbau
F Maschinenbau; Beleuchtung; Heizung; Waffen; Sprengen
G Physik
H Elektrotechnik



Patent Zeichnungen (PDF)

Copyright © 2008 Patent-De Alle Rechte vorbehalten. eMail: