Patent Application
20090235003
The present invention relates to devices and methods for memory control for performing effective memory access.
Recently, house-use LSIs are used in the form of unified memories including a single external memory in many cases in view of system cost down, so that the single memory receives various kinds of memory access requests. Further, with a plurality of functions installed, a higher bandwidth is demanded, needing high-speed memories.
Referring to a DRAM as one example, the operation frequency of the DRAM's memory cells has not been so changed as ever, and therefore, the minimum access size to the DRAM has been increasing more and more when viewed from the user's side. For this reason, though less or no problem is involved in transmitting data having long burst length, the amount of transmitted invalid data increases in transmitting data having short burst length to lower the effective bandwidth.
For example, in medium processing, a problem of lowering the effective bandwidth is involved in motion compensation necessitated for video coding. This problem could have been solved only by using an expensive DRAM that can withstand such lowering of the effective bandwidth (see, for example, Patent Document 1).
Nevertheless, as described above, when a DRAM exhibiting high performance in data transmission is employed, the amount of transmitted invalid data increases in transferring data having short burst length to lower the effective bandwidth. Further, when an access circuit accessible to a plurality of storage devices accesses one of the storage devices, the access request of the access circuit accessible to the plurality of storage devices is kept waiting if the storage device has already received an access request from another access circuit.
In addition, if a storage device out of the accessible storage devices receives no access request from any other access circuit, the bandwidth of this storage device is wasted by the wait time.
Consider next the case of data transmission, such as data copy or the like among a plurality of storage devices. One of access circuits accesses one of storage devices first, data which is stored in this memory and to which another access circuit is to access is stored into another storage device to which the other access circuit is accessible, and only thereafter, the other access circuit accesses the thus stored data. This data transmission, however, takes much time when processing a large amount of data. Storage devices to which an access circuit is accessible is usually used for another purpose, for example, as local memories or the like for storing data relating to the access circuit, and therefore, an additional memory area for data transmission between the plurality of storage devices must be reserved. If the storage devices cannot perform time sharing or the like, the capacities or the bandwidth of the memories must be increased or another countermeasure should be provided. An increase in memory capacities, memory bandwidths, or the like necessitates similar countermeasures by the number of master storage devices, increasing the circuit area.
Further, provision of an access circuit accessible to a plurality of storage devices complicates an arbitration circuit to thus increase the circuit area and power consumption. A plurality of such access circuits involves the similar problems of which number is equal to the number of the access circuits.
In the case where such an LSI is developed to a low-end field, a single storage device may suffice because the bandwidth request is low. In this case, however, all access circuits must be accessible to the single storage device. Such a configuration increases the circuit area for only development to a low-end field, involving wiring congestion in layout design of an LSI and the like.
The present invention has been made in view of the foregoing and has its object of enhancing an effective bandwidth.
To attain the above object, the present invention provides a memory control device including: at least two storage devices in which data is stored; at least two access means which access a storage device; and an arbitration circuit which arbitrates access requests issued from the respective access means for each of the storage devices.
With the above arrangement in accordance with the present invention, the amount of transmitted invalid data out of data having short burst length can be reduced, attaining advantageous effect in enhancing the effective bandwidth. Further, not every access circuit has to be accessible to the plurality of storage devices, leading to advantageous effect in reducing the circuit area.
Further, access to the storage devices in an efficient sequence is enabled to increase the effective bandwidth further in each storage device. Moreover, some of the access circuits needs not to be accessible to the plurality of storage devices, attaining advantageous effects in reducing the circuit area. As well, this is advantageous in reducing the circuit area even when taking development of the LSI into consideration and attains quick activation and advantageous effects in reducing power consumption.
Embodiments of the present invention will be described below in detail with reference to the accompanying drawings. The following preferred embodiments describe mare essential examples and does not intend to limit the present invention, applicable subjects, and use thereof.
Though
The respective arbitration circuits 20, 21 arbitrate access requests issued from the access circuits 30, 40 to the storage devices 10, 11 for each storage device 10, 11.
Each storage device 10, 11 stores necessary data and reads out data in response to an access request and is, specifically, composed of a DDR2 (Double Data Rate 2).
Suppose herein that the bus width of a data bus 500 between the arbitration circuit 20 and the storage device 10 is set at four bytes while the bus width of a data bus 501 between the arbitration circuit 21 and the storage device 11 is set at four bytes. Accordingly, the minimum access unit is four burst, namely, 16 bytes.
As a comparative example for comparing the performance of the memory control device in accordance with Embodiment 1, a configuration of a conventional memory control device is shown in
Suppose herein that the bus width of a data bus 502 between the arbitration circuit 22 and the storage device 12 is set at eight bytes and a DDR2 is employed as the storage device 12. Accordingly, the minimum access unit is four burst, namely, 32 bytes.
The amount of transferred invalid data will be examined below specifically. When supposing that the access circuits 30, 40 of the memory control circuit shown in
On the other hand, the access circuits 30, 40 of the conventional memory control device shown in
Further, when no access is required in the storage device 10 from the two access circuits 30, 40 of the memory control device of Embodiment 1, wait time required for arbitration in the arbitration circuit 20 is reduced in general when viewed from one of the access circuits, which is preferable.
It is noted that though the memory control device of Embodiment 1 employs DRAMs (Dynamic Random Access Memories) as the storage devices 10, 11 but the present invention is not limited thereto and may employ SRAMs (Static Random Access Memories), flash memories, or the like, for example.
The storage devices 10, 11 may be different in kind from each other. For example, the storage device 10 may be a DRAM while the storage device 11 may be a flash memory.
The memory control device of Embodiment 1 uses the two storage devices 10, 11 but may use two or more storage devices. The storage devices 10, 11 may have any bus widths.
The present embodiment refers to the access circuits 30, 40 each accessible to the storage devices 10, 11, but the access circuits 30, 40 may be accessible to only either storage device.
Furthermore, in the case where a circuit performing the operation of the memory control device in Embodiment 1 is composed of an LSI, the access circuits 30, 40 may be provided internally or externally.
As shown in
Between the two arbitration circuits 21, 21, an inter-storage-device transfer circuit 50 is provided for performing data transmission between the storage devices 10, 11.
As shown in
On the other hand, in the case where data in the storage device 11 to which the access circuit 40 accesses is required by the other access circuit 30, the inter-storage-device transfer circuit 50 copies, on the basis of a signal 1001 output from the access circuit 40, the necessary data from the storage device 11 to the storage device 10.
In this way, provision of the inter-storage-device transfer circuit 50 eliminates the need of the access circuits 30, 40 to access the plurality of storage devices 10, 11, leading to advantages in reducing the circuit area and power consumption and attaining data copy between the storage devices.
In the arbitration circuits 20, 21, if data is copied when there is no access, of which real time performance should be guaranteed, from the access circuits 30, 40, data copy using effective vacant bandwidth can be attained with real time performance of each access circuit 30, 40 guaranteed, thereby enhancing the operation efficiency.
The access circuits 30, 40 are accessible to a single storage device 10 or 11 in
The arbitration circuit 20 under the state where the storage device 10 is accessible outputs a signal 1010 indicating the access state to each access circuit 30, 40.
As well, the arbitration circuit 21 under the state where the storage device 11 is accessible outputs a signal 1011 indicating the access state to each access circuit 30, 40.
The access circuits 30, 40 access an optimum storage device on the basis of the respective signals 1010, 1011.
The above control enables immediate receipt of access from, for example, the access circuit 30 that has received the signal 1010, irrespective of the access state of the other access circuit 40.
In other words, if some storage device to which an access circuit accesses is busy because of access from another access circuit, access chance to another accessible storage device having low access frequency may be lost until the busy storage device becomes free. The above control, however, prevents such access change from being lost, which is advantageous.
To the primary storage section 70, a vacant information managing device 71 is connected which outputs access requests from the access circuits 30, 40 and receives pointer information indicating a data storage state of the primary storage device 70.
The vacant information managing device 71 compares the pointer information with a predetermined set value and informs the access circuit 30 through the signal 1010 about vacant information, as the signal 1010, of the primary storage device 70 according the comparison result.
The predetermined set value to be compared is preferably set with a time period taken into consideration from time when the vacant information is informed to the access circuit 30 to time when a command of an access request issued from the access circuit 30 reaches the arbitration circuit 20.
The arbitration section 80 performs arbitration of access requests from the access circuits 30, 40 and outputs the access request issued from a selected access circuit to the storage device 10.
Further, when the access circuit 30 becomes accessible by arbitration of the arbitration section 80, the arbitration section 80 issues the signal 1010 to inform the access circuit 30 about the accessibility.
For example, the signal 1010 indicating the vacant information may be output with taking into consideration timing that the access circuit 30 becomes necessarily accessible after several cycles, namely, a time period from time when the arbitration section 80 outputs the signal 1010 indicating the vacant information to the access circuit 30 to time when the arbitration section 80 receives an access request issued in the access circuit 30 on the basis of the signal 1010.
The primary storage devices 72, 73 may have any number of stages. The primary storage devices 72, 73 may not necessarily be provide to the access circuits 30, 40, respectively, and the access circuits 30, 40 may share a primary storage device.
The access circuit 40 is accessibly connected to the storage device 10 through the arbitration circuit 20 and is accessibly connected to the storage device 11 through the arbitration circuit 21.
The switching circuit 90 switches an access target of the access circuit 30 on the basis of a set value of a register 91, which will be described later, and specifically, switches between the storage devices 10, 11 to be accessed.
To the switching circuit 90, the register 91, which is externally accessible, is connected. The register 91 stores information indicating to which storage device an access request is to be accessed. Value setting of the register 91 changes the access target between the storage devices 10, 11.
The above configuration leads to advantages in reducing the circuit area and power consumption in the memory control device. Specifically, though the access circuit 30 accessible to both the storage devices 10, 11, will increase the circuit area and power consumption in general, application of the present invention to an access circuit which requires access only to the storage device 10, for example, in a given application leads to advantages in reducing the circuit area and power consumption.
The selector 100 outputs only one of access requests from the access circuits 30, 40 selectively to the storage device 10 through the arbitration circuit 20.
This configuration eliminates the need to provide a plurality of storage devices, and can be applied directly to the case, for example, where the same LSI is developed to a low-end field having low bandwidth request. As a result, wiring congestion in LSI design can be obviated with an increase in circuit area suppressed.
With the configuration, in which an output of a data arbitration circuit is selected for the storage device, the circuit area can be reduced and wiring congestion in layout design can be obviated.
Specifically, in the case where there are many access circuits, input wirings of the selector 110 increases to influence the circuit scale and to invite wiring congestion in layout design. Nevertheless, the above configuration of the memory control device of Embodiment 6 leads to advantages in solving this problem.
Further, the circuit source of the data arbitration circuits 25, 26 of the memory control device in Embodiment 6 is equivalent to those of the memory control device in Embodiment 1 even though no such high bandwidth request is demanded. Hence, the performance is further improved.
The arbitration circuit 20 is connected to the storage device 10 through the selector 110. The arbitration circuit 21 is connected to the storage device 11 and is connected to the storage device 10 through the selector 110.
To the arbitration circuit 21, a register 120 is connected which outputs to the arbitration circuit 21 a signal 1030 for controlling clock oscillation or stop.
Further, a register 121 is connected to the storage device 11. In the case, for example, where the storage device 11 is a DRAM, the register 121 outputs to the storage device 11 a signal 1031 for controlling power down or self-refresh mode activation or stop.
With the above configuration, when values of the registers 120, 121 are set in a standby mode in which almost all equipment's functions are stopped, the arbitration circuit 21 can be set in a clock stop state while the storage device 11 can be set in a power down state or in the self-refresh mode, thereby suppressing power dissipation.
On the other hand, the arbitration circuit 20 and the storage device 10 are in operation. If an instruction or data of a microcomputer or the like necessary for system recovery is stored in the storage device 10, it is unnecessary to develop the instruction or the data of the microcomputer to the storage device 10 again at recovery from the standby mode, attaining quick activation of the equipment.
As described above, the present invention attains highly practical effects that the effective bandwidth can be enhanced and is therefore much useful and highly applicable to industries. For example, the present invention can be applied to a network terminal reproducing a compressed and coded stream, a DVD recorder/player, a digital television set, a PDA, a mobile phone, a personal computer, and the like.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2005-169817 | Jun 2005 | JP | national |
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/JP2005/023786 | 12/26/2005 | WO | 00 | 12/6/2007 |
