Semiconductor device including multi-chip

Description Submit all comments and votes

BACKGROUND OF THE INVENTION

The present invention relates to a combined type semiconductor memory module, and in particular to a DRAM refresh method.

A list of documents referred to herein is as follows. The documents will be referred to by using a document number.

Document 1: LRS1337 Stacked Chip 32M Flash Memory and 4M SRAM Data Sheet (retrieved on Apr. 21, 2000), Internet <URL:http://www.sharpsma.com/index.html>

Document 2: JP-A-11-219984 (laid open in Aug. 10, 1998) (corresponding to U.S. Pat. No. 6,157,080 published in Dec. 5, 2000)

Document 3: JP-A-5-299616 (laid open in Nov. 12, 1993) (corresponding to EUROPEAN PATENT APPLICATION Publication number 566,306 laid open in Oct. 20, 1993)

Document 4: JP-A-8-305680 (laid open in Nov. 22, 1996)

Document 5: JP-A-11-204721 (laid open in Jul. 30, 1999)

Document 6: JP-A-10-11348 (laid open in Jan. 26, 1998)

In the document 1, there is described a combined type semiconductor memory module including a flash memory and an SRAM sealed integrally with a BGA (ball grid array) type package by using a stacked chip configuration. The flash memory and the SRAM share address input terminals and data input and output terminals with respect to input and output electrodes of an FBGA (Fine-pitch Ball Grid Array) type package. However, they have independent control terminals, respectively.

In the document 2, there is described a combined type semiconductor memory module including a flash memory and an SRAM integrally sealed to a BGA (ball grid array) type package by using a stacked chip. Signal pads of the flash memory are subject to face down bonding to a circuit substrate of the BGA package via solder bumps. Signal pads of the SRAM mounted on the flash memory are connected to the substrate by wire bonding.

With reference to FIG. 17 of the document 3, there is described a combined type semiconductor memory module including a flash memory chip and a DRAM chip integrally sealed to a lead frame type package. Furthermore, with reference to FIG. 1, there is described such a flash memory and a DRAM that address input terminals, data input and output terminals, and control terminals are shared for inputting and outputting with respect to input and output electrodes the package.

In the document 4, there is described a semiconductor device. In this semiconductor device, an SRAM chip is mounted on a die pad. On the SRAM chip, a flash memory chip and a microcomputer chip connected via a bump electrode are mounted. Those chips are sealed integrally with a lead terminal type package to form the semiconductor device.

With reference to FIG. 15 of the document 5, there is described such a semiconductor device that two small-sized chips are mounted on the back of one large-sized chip via an insulation plate and those chips are integrally sealed to a lead frame type package. It is described that there are a flash memory chip, a DRAM chip, and an ASIC (Application Specific IC) as a combination of chips that can be mounted and consequently a memory embedded logic LSI is implemented by one package.

In the document 6, there is described a technique for avoiding collision between access from the outside and refresh of the DRAM by providing two DRAM blocks, storing the same data in duplicate, and staggering two DRAM blocks in refresh timing. This control is conducted by a DRAM controller. This DRAM controller issues physically independent address signals and control signals to the two DRAM blocks.

SUMMARY OF THE INVENTION

Prior to the present invention, the present inventors examined a mobile phone and a combined type memory module that is to be used for the mobile phone and that includes a flash memory and an SRAM mounted on one package. Besides the OS (operation system) of the mobile phone system, a communication program and application programs are stored in the flash memory. On the other hand, telephone numbers, an address book, and a ringing tone are stored in the SRAM. Besides, a work area which is temporarily used at the time of execution of an application is secured in the SRAM.

In order to retain data to be stored, such as the telephone numbers and the address book, power supply for retaining the data is connected to the SRAM even when power supply of the mobile phone is in the off state. For retaining the data over a long period of time, it is desirable that the data retention current of the SRAM is small. However, it is expected that with an increase of the functions added to the mobile phone (downloading of music and games) the work area used by the applications becomes large and an SRAM having a larger storage capacity is needed. In particular, enhancement in the function of recent mobile phones is remarkable. It has been found that it is gradually becoming difficult to cope with the function enhancement by using an SRAM having a larger capacity. In other words, an increase in the capacity of the SRAM has the following problem. The problem of the large capacity SRAM is that the data retention current is increased by the amount of increase of the storage capacity and in addition the data retention current is increased by increase of the gate leakage current. The reason is as follows: if advanced scaling technologies are introduced and the oxide insulation films of MOS transistors are made thinner in order to implement large capacity SRAMs, then a tunnel current flows from the gate to the substrate and the data retention current is increased thereby.

Therefore, one of objects of the present invention is to implement a memory that is larger in storage capacity and smaller in data retention current.

According to one aspect of the present invention, a flash memory, a static random access memory (SRAM), and a dynamic random access memory (DRAM) having a plurality of memory banks are incorporated into one sealing member or package. In the dynamic random access memory (DRAM), reading/writing is conducted by a command synchronized to a clock. On the sealing member, there are provided electrodes for conducting wiring to semiconductor chips and electrodes for making connections between the sealing member and the outside of the sealing member.

In order to hide refresh of the DRAM from the outside of the semiconductor device at this time, a memory controller is connected to the DRAM including two or more banks in one chip. The memory controller controls memory access to the DRAM. In the case where memory access has been conducted by the memory controller in the first interval, the first bank may be accessed. In the case where memory access has been conducted by the memory controller in the second interval, the second bank may be accessed.

A dynamic random access memory (DRAM) having a plurality of memory banks is used. In the dynamic random access memory (DRAM), reading/writing is conducted by a command synchronized to a clock. A plurality of memory banks are assigned to a first memory block and a second memory block having the memory capacity. Memory access is conducted alternately in the first interval and the second interval. In the first interval, a read/write command directed to the DRAM is executed on the first memory block, and on the second memory block, refresh operation is executed preferentially. In the second interval, a read/write command directed to the DRAM is executed on the second memory block, and on the first memory block, refresh operation may be executed preferentially.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a configuration diagram of a memory module according to an embodiment of the present invention;

FIG. 2 is a block diagram showing an example of CHIP2 illustrated in FIG. 1;

FIGS. 3A and 3B are diagrams showing examples of an address map of a memory module according to an embodiment of the present invention;

FIGS. 4A and 4B are diagrams showing examples of an address map of a memory module according to an embodiment of the present invention;

FIG. 5 is a diagram showing an example of a configuration of an ATD circuit or a DTD circuit illustrated in FIG. 2;

FIGS. 6A and 6B are diagrams showing examples of a refresh system of a DRAM;

FIG. 7 is a flow chart showing a flow of processing conducted when a DRAM is accessed;

FIG. 8 is a flow chart showing a flow of operation in a DRAM bank conducted during a REF interval;

FIGS. 9A and 9B are diagrams showing how an access to a DRAM and its refresh are conducted simultaneously;

FIG. 10 is a block diagram showing a configuration example of a flash memory;

FIG. 11 is a block diagram showing a configuration example of an SRAM;

FIG. 12 is a block diagram showing a configuration example of a DRAM;

FIG. 13 is a diagram showing an example of a time chart of a memory module according to an embodiment of the present invention;

FIG. 14 is a block diagram showing a configuration example of CHIP2 illustrated in FIG. 1;

FIG. 15 is a diagram showing an embodiment of a large capacity memory of an unsynchronous SRAM interface system utilizing a DRAM according to an embodiment of the present invention;

FIGS. 16A and 16B are diagrams showing an example of a mounting form of a memory module according to an embodiment of the present invention;

FIGS. 17A and 17B are diagrams showing an example of a mounting form of a memory module according to an embodiment of the present invention;

FIGS. 18A and 18B are diagrams showing an example of a mounting form of a memory module according to an embodiment of the present invention;

FIGS. 19A and 19B are diagrams showing an example of a mounting form of a memory module according to an embodiment of the present invention;

FIGS. 20A and 20B are diagrams showing an example of a mounting form of a memory module according to an embodiment of the present invention;

FIGS. 21A and 21B are diagrams showing an example of a mounting form of a memory module according to an embodiment of the present invention;

FIGS. 22A and 22B are diagrams showing an example of a mounting form of a memory module according to an embodiment of the present invention; and

FIGS. 23A and 23B are diagrams showing an example of a mounting form of a memory module according to an embodiment of the present invention.

DESCRIPTION OF THE EMBODIMENTS

Hereafter, embodiments of the present invention will be described in detail by referring to the drawing. Circuit elements forming respective blocks of the embodiment are not especially limited, but they are formed on one semiconductor substrate, such as monocrystalline silicon, by a known integrated circuit technique of CMOS (complementary MOS).

<Embodiment 1>

FIG. 1 shows a first embodiment of a memory module that is an example of a semiconductor integrated circuit device according to an embodiment of the present invention. The present memory module is formed of three chips. Hereafter, respective chips will be described.

First, CHIP1 (FLASH) is a nonvolatile memory. A ROM (read only memory), an EEPROM (electrically erasable and programmable ROM), or a flash memory can be used as the nonvolatile memory. The present embodiment will be described by taking a flash memory as an example. A static random access memory (SRAM) and a control circuit (CTL_LOGIC) are integrated in CHIP2 (SRAM+CTL_LOGIC). The control circuit controls the SRAM integrated in the CHIP2 and CHIP3. The CHIP3 (DRAM) is a dynamic random access memory (DRAM). As the DRAMs, there are various kinds, such as EDO, SDRAM, and DDR, depending upon differences in internal configuration and interface. Any DRAM can be used in the present memory. However, the present embodiment will be described by taking an SDRAM as an example.

From the outside, address signals (A0 to A20) and command signals (S-/CE1, S-CE2, S-/OE, S-/WE, S/LB, S-/UB, F-/WE, F-/RP, F-/WP, F-RDY/BUSY, F-/CE, and F-/OE) are inputted to this memory module. Power supply is fed through S-VCC, S-VSS, F-VCC, F-VSS, L-VCC and L-VSS. I/O0 to I/O15 are used for inputting and outputting data. Address signal lines and data input/output lines are connected to CHIP1 (FLASH) and CHIP2 (SRAM+CTL_LOGIC) in common. The CHIP2 supplies a clock (D-CLK), an address (D-A0 to D-A13), a command (D-CKE, D-/CS, D-/RAS, D-/CAS, D-/WE, D-DQMU/DQML), data for DRAM (D-DQ0 to D-DQ15), and power supplies (D-VCC, D-VSS, D-VCCQ, D-VSSQ). One of features of the input/output nodes between this memory module and the outside is that signal terminals for DRAM interface are made directly invisible. While a BGA (ball grid array) type package of the present invention will be described later with reference to FIG. 16 showing an embodiment, therefore, signal terminals for controlling the DRAM are typically not provided in terminals used as external terminals in this package. As a result, the existence of the DRAM is hidden, and the merit of an increase of the storage capacity can be given. The user need not take DRAM refresh into consideration. In the case where the number of signal terminals of the BGA type package is very large and enough to spare, however, control terminals of the DRAM may be pulled out to the outside in parallel mainly for the purpose of test conducted by a manufacturer at the time of fabrication. By doing so, it becomes possible for the manufacturer to conduct defect analysis rapidly. As a matter of course, this function is not usually opened to the user.

Respective command signals will now be described briefly. S-/CE1 and S-CE2 inputted to the CHIP2 are chip enable signals. S-/OE is an output enable signal. S-/WE is a write enable signal. S-/LB is a lower byte selection signal. S-/UB is an upper byte selection signal. F-/WE inputted to the CHIP1 is a write enable signal. F-/RP is a reset/ deep power down signal. F-/WP is a write protect signal. F-RDY/BUSY is a ready/busy output signal. F-/CE is a chip enable signal. F-/OE is an output enable signal and is used to control the flash memory.

In the present memory module, it is possible to access the flash memory, SRAM and DRAM by using the common address lines (A0 to A20) and data input/output lines (I/O0 to I/O15).

When accessing the flash memory (CHIP1), required signals among the command signals F-/WE, F-/RP, F-/WP, F-RDY/BUSY, F-/CE, and F-/OE are made active besides the address lines (A0 to A20). When accessing the SRAM (CHIP2) or DRAM (CHIP3), required signals among the command signals S-/CE1, S-CE2, S-/OE, S-/WE, S-/LB and S-/UB are made active besides the address lines (A0 to A20). Either access is conducted by using the so-called SRAM interface system.

Access to the SRAM and access to the DRAM are discriminated by the address value. On the basis of the inputted address value, the control circuit (CTL_LOGIC) judges the access destination. The range of address that becomes an access to the SRAM and the range of address that becomes an access to the DRAM are determined by previously setting a value in a register provided in the control circuit (CTL_LOGIC).

When accessing the DRAM, the control circuit (CTL_LOGIC) generates an address signal and command signals required for accessing the DRAM, and accesses the DRAM. In the case of a read access, data read out from the DRAM is read out from data I/O for DRAM (D-DQ0 to D-DQ15) temporarily into the control circuit (CTL_LOGIC), and then outputted to data input/output lines (I/O0 to I/O15) of the memory module. In the case of write access, write data is inputted from the data input/output lines (I/O0 to I/O15) of the memory module, and then inputted to the DRAM through the data I/O for DRAM (D-DQ0 to D-DQ15).

Power supply to the DRAM is fed from the L-VCC and L-VSS, and connected to D-VCC, D-VSS, D-VCCQ and D-VSSQ through the control circuit (CTL_LOGIC). Supply of power to the DRAM is controlled by a command signal PS, and it can be disconnected as occasion demands. When throwing in the disconnected DRAM power again, it is necessary to initialize the DRAM. The signal generation and timing control required to initialize the DRAM is conducted by the control circuit (CTL_LOGIC).

Furthermore, when refreshing the DRAM, the control circuit (CTL_LOGIC) can conduct it by periodically issuing a refresh command. In general, the refresh characteristics of the DRAM are aggravated when the temperature is high. By providing a thermometer in the control circuit (CTL_LOGIC) and narrowing the issue period of the refresh command, the DRAM can be used in a wider temperature range. On the contrary, when the temperature is low, power required for data retention can be reduced by widening the issue period of the refresh command.

In addition, the control circuit (CTL_LOGIC) retains one data in two different addresses of the DRAM and adjusts timing of conducting refresh. Thus the control circuit (CTL_LOGIC) hides the refreshing from the outside of the DRAM so as not to restrict access by the refresh operation.

According to the embodiment heretofore described, a large capacity memory module using an inexpensive general purpose DRAM can be implemented while following the SRAM interface system. In the memory module according to the present invention, a DRAM is used. Since refreshing required for the DRAM is executed within the module, however, the DRAM can be used without considering the refresh in the same way as the SRAM. Furthermore, by changing the refresh period executed within the module according to the temperature, it becomes possible to widen the use temperature range of the DRAM and reduce the power required for data retention. A large capacity memory module having a wide use temperature range can be implemented.

In addition, the refresh of the DRAM can be hidden from the outside of the DRAM by duplicating the data retention in the DRAM and adjusting the refresh timing. When accessing the present memory module, therefore, there is no need to adjust timing while taking the refresh into consideration. Accordingly, the present memory module can be used in the same way as the conventional memory module using only the SRAM. Without changing the conventional system, therefore, a large capacity memory module can be used.

Another object of the present invention is to implement a memory module having a small data retention current. This object can be achieved by disconnecting power supplied to the DRAM and retaining only data stored in the SRAM. By storing only data to be retained in the SRAM and stopping power supply to a memory for storing data that need not be retained, it is possible to retain only required data with a minimum data retention current.

FIG. 2 shows the CHIP2 (SRAM+CTL_LOGIC). The CHIP2 (SRAM+CTL_LOGIC) is formed of the SRAM and the control circuit (CTL_LOGIC). The integrated SRAM is an unsynchronous SRAM typically used heretofore. The control circuit (CTL_LOGIC) is a portion of the CHIP2 other than the SRAM and is shown as an area surrounded by a broken line. The control circuit (CTL_LOGIC) includes AS, MMU, ATD, DTD, FIFO, R/W BUFFER, A_CONT, INT, TMP, RC, PM, CLK_GEN and COM_GEN.

When an address is inputted from the outside, a memory management unit MMU converts the inputted address according to a preset value, and selects a memory to be accessed. If the SRAM is selected, then a command signal is sent to the SRAM by an access switch (AS) and the SRAM is accessed. An address transition detector circuit (ATD) detects changes in the address signal and the command signals, and outputs a pulse. A data transition detector circuit (DTD) detects changes in the data signal and the command signals, and outputs a pulse. A R/W BUFFER retains data temporarily for DRAM reading and writing. A first-in first-out memory FIFO is a first-in first-out buffer circuit, which temporarily retains data written into the DRAM and its address. An initializing circuit INT initializes the DRAM when power supply to the DRAM is started. A temperature measuring module (TMP) detects the temperature, and outputs a signal depending on the detected temperature to the RC and A_CONT. The RC is a refresh counter, which generates an address of refreshing according to the refresh period of the DRAM. Furthermore, the refresh counter RC alters the refresh period according to the temperature in response to an output signal of the temperature measuring module (TMP). A power module (PM) conducts power supply to the control circuit (CTL_LOGIC) of the CHIP2 and the DRAM and control of the power supply. A clock generator (CLK_GEN) generates a clock and supplies it to the DRAM and the control circuit (CTL_LOGIC). A command generator (COM_GEN) generates a command required to access the DRAM. An access controller (A_CONT) conducts control of the whole operation of the CHIP2 (SRAM+CTL_LOGIC) and generates an address for accessing the DRAM.

For conducting memory access to the CHIP2 (SRAM+CTL_LOGIC), interfacing is executed by using the unsynchronous SRAM system, which has been heretofore used in general. When an address (A0 to A21) is inputted to the CHIP2 (SRAM+CTL_LOGIC) from the outside, the address value is first converted by the MMU. The pattern of conversion is determined by a value previously inputted to a register within the MMU. It is determined on the basis of the converted address whether the access destination is SRAM or DRAM.

When accessing the SRAM, the MMU sends the converted address to the SRAM. At the same time, the MMU orders the address access switch (AS) to transfer the command. The address access switch (AS) transfers the command to the SRAM, and access to the SRAM is started. As for the ensuing operation, access to the so-called unsynchronous SRAM is conducted.

Operation of respective blocks of the control circuit conducted when executing read access to the DRAM will hereafter be described. First, the address inputted from the outside and converted by the MMU and the command sensed by the ATD are sent to the A_CONT. The A_CONT judges the execution of the access to the DRAM on the basis of the sent address and command, and orders the COM_GEN to issue a command to the DRAM. Furthermore, the A_CONT converts the address received from the MMU to an address for the DRAM, and outputs the address for the DRAM to the DRAM. In synchronism with the clock generated by the CLK_GEN, the COM_GEN issues a command to the DRAM. Upon receiving the command and the address, the DRAM outputs data. The outputted data is transferred to the I/O0 to I/O15 via the R/W BUFFER. The read access is thus finished.

When executing write access to the DRAM, the address inputted from the outside and converted by the MMU, the command sensed by the ATD, the command and data sensed by the DTD are sent to the A_CONT. The A_CONT judges execution of access to the DRAM on the basis of the sent address and command, and orders the COM_GEN to issue a command to the DRAM. Furthermore, the A_CONT converts the address received from the MMU to an address for the DRAM, and outputs the address for the DRAM to the DRAM. In synchronism with the clock generated by the CLK_GEN, the COM_GEN issues a command to the DRAM. Data to be written is inputted from the I/O0 to I/O15 and temporarily retained in the R/W BUFFER, then sent to the DRAM and written into the DRAM. In addition, the data thus written and the address are retained in the FIFO as well, and written into a different bank of the DRAM as well later.

Power supplied to the DRAM is controlled by the power module (PM). In some cases, it is desirable to reduce the consumption current of the device having the memory module mounted thereon according to the operation state. In such a case, the power module can stop the refresh conducted by the refresh counter according to, for example, the command signal PS, and thereby reduce power required to refresh the DRAM. In the case where it is desirable to further reduce the consumption power, power supplied to the DRAM can be disconnected within the memory module. In this case, the power module stops power to the D-VCC supplied to the DRAM according to the command signal PS outputted by the device.

In the case where it is desirable to further reduce the power consumption, the power module can stop supply of power to a portion of the CHIP2 (SRAM+CTL_LOGIC) concerning the memory access to the DRAM, according to the command signal PS. In this state, it is possible to connect power supply to, for example, only the MMU and AS of the CHIP2 (SRAM+CTL_LOGIC) besides the SRAM, bring the MMU and AS into the operation state, and bring about such a mode as to execute only access to the SRAM.

In addition, it is also possible to bring about such an operation state as to conduct only date retention of the SRAM by the command PS. In such a case, power supplies other than power supplies (S-VCC and S-VSS) connected to the SRAM are disconnected and access to a memory is inhibited. In this state, the memory module retains data stored in the SRAM.

For re-activating the DRAM stopped in operation by temporarily stopping the power supply, it is necessary to initialize the DRAM in addition to resuming the power supply. The initializing method is typical. In the memory module, however, the initializing circuit (INT) orders the access controller (A_CONT) to execute an initializing procedure, and initialization is executed.

Also in the case where the refresh of the DRAM is stopped, initialization of the DRAM is required for making the DRAM operate again. In this case as well, the initializing circuit (INT) directs an initializing procedure to the access controller (A_CONT) and initialization is executed.

The refresh counter RC outputs a refresh address in accordance with the refresh period of the DRAM, and requests the access controller to execute the refresh. In response to the request from the refresh counter, the access controller issues a refresh command while arbitrating with respect to the access to the DRAM from the outside, and conducts the DRAM refresh.

When using the memory module in a high temperature state, it becomes necessary to shorten the refresh period of the DRAM and conduct the refresh frequently. In such a case, the temperature measuring module (TMP) detects the temperature, and notifies the temperature to the refresh counter and the access controller. When the temperature becomes high, the refresh counter changes the refresh period so as to make it shorter, and outputs the refresh address.

The clock (D-CLK) required for the operation of the DRAM is generated by the clock generator (CLK_GEN). The clock generator supplies the clock to respective blocks within the control circuit as well as the DRAM. In the case where the DRAM operates in synchronism with the clock, the command issue of the command generator (COM_GEN) is conducted in synchronism with the clock.

By stopping the refresh of the DRAM, the power consumption can be reduced. In the case where the power of the DRAM is disconnected and only the SRAM is accessed, operation with low power becomes possible although the storage capacity is small. In this case, it is also possible to further stop the power supplied to the control circuit required for DRAM access and thereby implement operation with further low power. Furthermore, by supplying power to only the SRAM and retaining only the data stored in the SRAM. Also when re-throwing in power to the DRAM, the DRAM initialization can be conducted by the control circuit, and consequently it is not necessary to execute the initialization procedure on the module from the outside. As a result, a memory module that reduces the power consumption can be implemented simply.

FIGS. 3A and 3B show examples of a memory map converted by the MMU. The present embodiment will now be described by taking a memory module having a storage area of 32 Mb in the nonvolatile memory, a data retention area of 2 Mb in the SRAM, and a storage area of 32 Mb in the DRAM, which is not restrictive, as an example. The address A0 to A20 inputted from the outside is used in common by the flash memory (CHIP1) and the CHIP2. For selection of the access destination, signals S-CS and F-CS for chip selection are used. If the signal F-CS has become active, then the CHIP1 is selected and access is conducted. If the signal S-CS has become active, then the CHIP2 is selected and access is conducted. The F-CS is a general term of the command signals F-/WE, F-/RP, F-/WP, F-RDY/BUSY, F-/CE and F-/OE. The S-CS is a general term of the command signals S-/CE1, S-/CE2, S/OE, S-/WE, S-/LB and S-/UB. If the CHIP2 is selected as the access destination, then the MMU selects a memory to be accessed, according to the address.

In the example of the memory map shown in FIG. 3A, the SRAM area is set so as to be concentrated to a part of an address space. The address space of the SRAM overlaps in the address space of the DRAM. Access to the overlapping address space is conducted for the SRAM. The DRAM area existing in the same address space becomes a shadow area, which is not accessed.

On the other hand, in the example shown in FIG. 3B, SRAM areas are set so as to be distributed into a plurality of address spaces. In this case as well, the address space of the SRAM overlaps in the address space of the DRAM. Access to the overlapping address space is conducted for the SRAM. In this example, the SRAM areas are set by taking 512 kb as the unit. This is set equal to the writing or erasing unit of the FLASH memory. This aims at facilitating handling conducted by the OS and a program by aligning the management unit of the address space with that of the FLASH memory.

In this way, the MMU can assign the SRAM areas and DRAM areas to specified address spaces. Especially when it is desirable to reduce the data retention current, it is possible to assign address spaces for storing data to be retained to SRAM-areas and stop power supply to the DRAM. Owing to this method, a memory module requiring a small data retention current can be implemented.

FIGS. 4A and 4B show different examples of the memory map converted by the MMU.

In the example of the memory map shown in FIG. 4A, the SRAM area is set so as to be concentrated to a part of an address space. The example of the memory map shown in FIG. 4A differs from that shown in FIG. 3A in that there is no overlapping between the address space of the SRAM and the address space of the DRAM. Since a shadow area is not generated in the DRAM, the memory space of the DRAM can be utilized efficiently. In FIG. 4B as well in the same way, there is no overlapping between the address space of the SRAM and the address space of the DRAM, unlike the example shown in FIG. 3B. Since the shadow area is not generated in the DRAM, the memory space of the DRAM can be used effectively. By using the memory map shown in FIG. 4A or 4B, the address space increases by approximately 2 Mb. This can be coped with by adding an address line A21. In this way, the storage area of the DRAM can be used more effectively in the memory map shown in FIG. 4A or 4B.

FIG. 5 shows a configuration example of the ATD circuit and its operation waveform. The address transition detector circuit (ATD) senses a value change of address signal lines and generates a pulse. Each of characters D1 and D2 used in the circuit diagram denotes a delay element for generating a delay. If a change occurs in address lines (A0 to AN), then the ATD outputs a pulse (/.phi.A0 to /.phi.AN) having a width equivalent to the sum of the delay of the delay element D1 and the delay of the delay element D2. Furthermore, by taking operation dispersion of individual address lines into consideration and generating a signal/.phi.ATD, which is obtained by adding these pulses, it is sensed that an address value appearing on the address lines has changed. As shown in FIG. 2, not only the address lines but also the command signals are connected to the ATD. The ATD thus detects that a new command has been inputted. The data transition detector circuit (DTD) has a configuration similar to that of the ATD. The DTD detects a change on the data lines and a command signal for writing, and recognizes write data and a write command.

Upon thus detecting unsynchronously changing SRAM interface signals by the ATD and DTD, the memory module starts its operation. By using these circuits, a memory module that operates on the basis of the unsynchronous SRAM interface can be implemented. Since unsynchronously changing signals pulsed and synchronized to the clock are sensed, it is also possible to use a memory device that conducts synchronous operation within the memory module.

FIGS. 6A and 6B show how the DRAM operates in time division in order to hide the refresh of the DRAM. By taking such a DRAM that one chip is formed of four banks, as an example, the DRAM operation will now be described. Four banks BANK-A0, BANK-A1, BANK-B0 and BANK-B1 are divided into two sets BANK-A0 and BANK-A1, and BANK-B0 and BANK-B1. The two sets are mapped to the same address space. In other words, a memory cell specified by one address exists in each of the two sets, and data are stored double redundantly.

FIG. 6A shows the operation of the DRAM when the temperature is below 75.degree. C., which is the typical use temperature range of the DRAM. Typically, memory cells of the DRAM need to be refreshed once every 64 ms. The 64 ms is divided into 8 intervals each having a length of 8 ms. The set of the BANK-A0 and BANK-A1 and the set of the BANK-B0 and BANK-B1 alternately operate interval after interval. A WORK interval represented as WORK in FIG. 6A represents an interval that the bank set operates. During the first WORK interval, the set of the BANK-A0 and BANK-A1 is operating.

If read access to the DRAM has been conducted, then readout is conducted from the set of the BANK-A0 and BANK-A1, which is in the WORK interval. If write access to the DRAM has been conducted, then writing into the set of the BANK-A0 and BANK-A1, which is in the WORK interval, is conducted and data thus written and the address are temporarily stored in the FIFO. The stored data is written into the set of the BANK-B0 and BANK-B1 as well in an interval T2. The interval T2 will be described later. While the BANK-A0 and BANK-A1 are in the WORK interval, the BANK-B0 and BANK-B1 are in a REF interval. In the REF interval, refresh is conducted for half an area of banks of the set of the BANK-B0 and BANK-B1. The REF interval is divided into a T1 interval and the T2 interval. During the T1 interval, refresh is conducted consecutively. During the T2 interval, data written during the REF interval is written back from the FIFO.

Assuming that the time required for the refresh is 70 ns per once, the time required for the refresh becomes 70 ns.times.2048 times, which is equal to 0.144 ms. Therefore, the interval T2 becomes 7.856 ms (=8 ms-0.144 ms). It is now assumed that the memory module is accessed once every 75 ns. If it is assumed that every access conducted during the REF interval is write access, then the maximum number of times thereof is 106,667 (=8 ms/75 ns). The time required for writing this into the DRAM is 7.47 ms (=106,667.times.70 ns), which is shorter than the T2 interval (7.856 ms). Even if refresh is conducted during the T1 interval, therefore, all of the write access conducted during the REF interval can be written back during the T2 interval.

Furthermore, it is also possible to execute the refresh in two banks simultaneously during the REF interval. In this case, the number of times of refresh executed in one bank during the T1 interval becomes the half, i.e., 1,024. If the T1 interval is shortened, then the storage capacity of the FIFO can be reduced, and in addition the period of access from the outside can be made shorter. A fast memory can thus be implemented.

FIG. 6B shows the case where the refresh period of the DRAM is reduced to half. In general, the refresh characteristic of the DRAM is aggravated when the temperature is high. When the temperature is high, for example, above 75.degree. C., therefore, data can be retained by shortening the refresh period as illustrated. In the present embodiment, the temperature is detected by the temperature measuring circuit TMP, and the refresh period is altered by the refresh counter and the access control circuit (A_CONT).

In this example, the refresh period of 64 ms is shortened to a half, which is 32 ns. Each of the WORK and REF intervals is 4 ms. If it is now assumed in the same way that every access conducted during the REF interval is write access, then the maximum number of times thereof is 53,334 (=4 ms/75 ns). The time required for writing this into the DRAM is 3.74 ms (=53,334.times.70 ns), which is shorter than the T2 interval (3.856 ms). Even if refresh is conducted during the T1 interval, therefore, all of the write access conducted during the REF interval can be written back during the T2 interval.

In this way, the refresh of the DRAM can be hidden. Though a general purpose DRAM is used, it is possible according to the present embodiment to hide the refresh and handle the DRAM in the same way as the unsynchronous SRAM. Therefore, it is possible to implement a large capacity memory module that can be accessed by the unsynchronous SRAM interface. Furthermore, in the case where the DRAM is used also when the temperature is high, the large capacity memory module can be implemented by only shortening the refresh period as in the present embodiment. On the other hand, when the temperature is low, power required for data retention can be reduced by widening the refresh period. While the operation unit of the DRAM is set equal to two banks in the present embodiment, it may be altered according to the configuration of the memory module or memory chip. Furthermore, while the refresh period of 64 ms is divided into eight intervals and each interval is made the WORK interval or REF interval, the storage capacity of the FIFO retaining the data and address can be reduced by further dividing the refresh period into shorter intervals. On the contrary, if the refresh period is divided into longer intervals, then the number of times of switchover from the WORK interval to the REF interval and vice versa can be reduced, and consequently the control circuit for the switchover can be formed simply.

FIG. 7 is a flow chart showing access to the CHIP3 (DRAM). In STEP1, an address is inputted and operation is started. In STEP2, the kind of access is judged on the basis of a command. Ensuing operation differs depending upon the access kind. If the access is readout, then the processing proceeds to STEP3. In the STEP3, data is read out from a bank that is in a WORK interval and operation is finished. If the access is write, the processing proceeds to STEP4 and STEP5. In the STEP4, data is written into a bank that is in a WORK interval. On the other hand, in the STEP5, written data and its address are retained in the FIFO. Upon transition of a bank that is in a REF interval from a T1 interval to a T2 interval, the processing proceeds to STEP6 and data retained in the FIFO is written into the bank that is in the REF interval.

By thus reading/wtiting data from/into the DRAM, the influence of the refreshing can be avoided. By using a large capacity DRAM, therefore, a memory module having an unsynchronous SRAM interface can be formed.

FIG. 8 is a flow chart showing operation of a bank that is in a REF interval of the CHIP3 (DRAM). STEP1 to STEP3 belong to a T1 interval, and STEP4 to STEP6 belong to a T2 interval. At the STEP1, the REF interval starts and refres