Memory unit and memory module using the same

1. Field of the Invention

The present invention relates to cache memory, and more particularly, the invention relates to a memory unit and a memory module applicable to a cache memory.

2. Description of the Related Art

Cache memory is located functionally between a processor and a main memory. The cache memory is generally faster but smaller than the main memory. The cache memory decreases memory access time by storing copies of portions of the contents of main memory. When a local process performs certain “cachable” operations requiring memory access, the cache memory is checked first to determine if it contains a copy of the information to be accessed. If the cache memory contains a copy of the information to be accessed, the processor performs the operation using the cache memory without accessing main memory. A cache “hit” occurs when the information to be accessed by a cachable operation is stored in the cache memory, and a cache “miss” occurs when the information is not.

A main memory address consists of two fields (not shown) The higher order bits are called the tag field, and the remaining lower order bits are called the index field. As shown in FIG. 1 , the cache memory 10 uses a tag memory, separate from data memory, to store the tag fields of the addresses of the main memory contents presently stored. When the processor performs memory access, the index field of the address addresses the tag memory. The output data from the tag memory is then compared bit by bit with the tag field of the main memory address. If the two fields are identical, a hit has occurred, and the corresponding data is supplied to the processor. If the fields are not identical, it is a miss, and main memory must be accessed.

If the contents of the tag memory become corrupted, memory accesses that should miss actually hit. This results in incorrect data supplied to the processor, a dangerous situation. It is common to use a signal parity bit over each byte in the tag memory, whereas, if a bad parity is detected when the tag memory is read, a miss condition is forced.

Conventionally, all information in tag memory is stored by SRAM cells, each having structure as shown in FIG. 2 a . During flushing operation, the tag memory is invalidated entry by entry and requires sequential circuits such that the performance of the cache system is affected.

Another conventional method stores the parity bit (control field) of the tag memory using memory banks with new cell structures as shown in FIGS. 2 b and 2 c , and conventional SRAM cells as shown in FIG. 2 a store information of the address field, attribute field and so on. Virtual ground/power terminals (VG/VP) of all cells for the parity bit in the tag memory are connected together to be forced ground/power voltage for flush operations. Although this method requires fewer additional control logic circuits, it requires asynchronous circuits and is poorly compatible with timing analysis. Further, this method has no tolerance for that there cannot be any signal glitch or system noise between the virtual power/ground terminals and the memory units, with which the data stored in the memory units may be unreliable.

It is therefore an object of the present invention to provide a memory unit with a simple structure to store the parity bit (control field) of the tag memory.

Another objection of the present invention is to provide a memory unit acting as a real synchronous element and is less sensitive to the signal glitch and system noise.

According to the above mentioned object, the present invention provides a memory unit. In the memory unit, a first access transistor and a second access transistor are coupled to a bit line pair. The first access transistor has a first terminal coupled to one bit line pair, and the second access transistor has a first terminal coupled to the other. A latch node is coupled between second terminals of the first access transistor and the second access transistor to latch data. A selection unit has two input terminals coupled to a word line and a flush line, and an output terminal is coupled to gates of the first access transistor and the second assess transistor. Invalidation information is written into the latch node from the bit line pair according to activations of the word line or the flush line.

The present invention also provides a memory module. In the memory module, at least one first memory region has a plurality of memory units. In each memory unit, the first access transistor has a first terminal coupled to one of a bit line pair, and the second access transistor has a first terminal coupled to the other. A latch node is coupled between second terminals of the first access transistor and the second access transistor to latch data. A selection unit has a first input terminal coupled to a word line, an output terminal coupled to gates of the first access transistor and the second assess transistor, and a second input terminal. The second input terminals of the selection units in all memory units are coupled to a flush line. Invalidation information is written into the latch nodes in the memory units from the bit line pair when the flush line is activated during a flush.

The present invention can be more fully understood by the subsequent detailed description and examples with reference made to the accompanying drawings, wherein:

FIG. 1 is a diagram of a cache memory;

FIG. 2 a shows a conventional memory unit;

FIG. 2 b shows another conventional memory unit;

FIG. 2 c shows another conventional memory unit;

FIG. 3 is a diagram of the memory module according to the present invention;

FIG. 4 is a diagram of the memory units in the memory module according to the present invention;

FIG. 5 shows the structure of the memory unit according to the present invention; and

FIG. 6 is a flowchart of the fabrication procedure of a cache memory according to the present invention.

FIG. 3 is a diagram of the memory module 20 according to the present invention. The memory module 20 includes two memory regions M 1 and M 2 , a driving buffer 22 , and a D-type Flip-Flop 24 . The memory region M 1 has a plurality of memory units storing the address field and attribute field in a tag memory. In the memory region M 1 , all memory units are convention SRAM cells as shown in FIG. 2 a . The memory region M 2 also includes a plurality of memory units storing entry control bits of the control field in a tag memory. The entry control bits also can be tag validation bits in a cache memory, tag modified/dirty bit in a copy-back/write-back cache memory and the like. FIG. 4 is a diagram of the memory region M 2 . In the memory region M 2 , the memory unit MU 1 is coupled to the word line WL 1 , the memory unit MU 2 is coupled to the word line WL 2 and so on. Further, all memory units MU 1 ˜MUn are coupled to a bit line pair (BL and {overscore (BL)}), and a flush line FC. The D-type Flip-Flop (DFF) 24 synchronizes a flush signal FCS by a clock signal CLK. The driver buffer 22 is coupled to the DFF 24 to activate the flush line when receiving the flush signal FCS from the DFF 24 . In the embodiment, the driving buffer 22 can include buffers connected in series (not shown).

The structure of each memory unit (MU 1 ˜MUn) is shown in FIG. 5 . As shown in FIG. 5 , each memory unit has a first access transistor T 1 , a second access transistor T 2 , a latch node L 1 and a selection unit OR 1 . In the present invention, the selection unit OR 1 can be an OR gate. The selection unit OR 1 has two input terminals coupled to a word line WL and the flush line FC, and an output terminal coupled to the gates of the first access transistor T 1 and the second access transistor T 2 . The first access transistor T 1 has a first terminal coupled to the bit line {overscore (BL)}, and a second terminal coupled to one end of the latch node L 1 . The second access transistor T 2 has a first terminal coupled to the bit line BL and a second terminal coupled to the other end of the latch node L 1 . In the present invention, the latch node L 1 has a first inverter INV 1 and a second inverter INV 2 . The inverter INV 1 has an input terminal coupled to the second terminals of the second access transistors T 2 , and an output terminal coupled to the second terminal of the first access transistors T 1 . The inverter INV 2 has an input terminal coupled to the second terminals of the first access transistors T 1 , and an output terminal coupled to the second terminal of the second access transistors T 2 .

During normal operation, the flush line FC is inactive and invalidation information can be written into the latch nodes L 1 in the memory unit MU 1 ˜MUn from the bit line pair (BL and {overscore (BL)}) according to activation of the corresponding word lines, and the memory unit acts as a conventional memory unit as shown in FIG. 1 a . For example, when the word line WL 1 is activated, the access transistors T 1 and T 2 are both turned on such that invalidation information is written to the latch node L 1 in the memory unit MU 1 from the bit line pair (BL and {overscore (BL)}). Also, when the word line WL 2 is activated, the access transistors T 1 and T 2 are both turned on such that invalidation information is written to the latch node L 1 in the memory unit MU 2 from the bit line pair (BL and {overscore (BL)}), and so on.

During a flush, all latch nodes L 1 in the memory unit MU 1 ˜MUn need to be invalidated at once regardless of activation of the word lines WL 1 ˜WLn. At this time, the driving buffer actives the flush line FC by the flush control signal FCS from the DFF 24 , such that invalidation information is written to the latch nodes L 1 in the memory unit MU 1 ˜MUn from the bit line pair (BL and {overscore (BL)}) when the flush line FC is activated.

A conventional method uses memory units, as shown in FIG. 2 b and FIG. 2 c , to store entry control bits in cache memory, and to change the state of the latch node by controlling the virtual ground/power connected to the source/drain side when the word lines are inactive during flush. Thus, in this conventional method, the memory units are asynchronous elements, and are sensitive to signal glitch, system noise, manufacture process and the like. In the present invention, during flush operation, however, the state of the latch nodes is changed when the word lines and flush lines are activated, such that the memory units in the present invention are real synchronous elements and less sensitive to signal glitch, system noise, manufacture process and so on. Further, the present invention is well suitable to timing analysis, since the state of the latch nodes is changed by controlling the gate side of the access transistor without drain/source side. Moreover, in the present invention, the memory module 20 is applicable to direct-mapping storage, set-associative mapping storage and full-associative mapping storage.

The present invention further provides a fabricating procedure for a cache memory, as shown in FIG. 6 . This fabricating procedure for a cache memory is described as follows.

First, in step S 100 , a specification for a desired cache memory is determined. For example, the specification may include entry number requirement, speed requirement, performance requirement and the like. Next, in step S 200 , conditions of a tag memory in the desired cache memory is determined. In this case, the conditions include at least the size and flush cycle time of the tag memory and the like. In step 300 , the tag memory is then implemented as an IP module or an object according to the conditions determined in step S 200 . The tag memory implemented to store entry control bits of the control field in a tag memory uses a plurality of memory units with structure as shown in FIG. 5 . The entry control bit can also includes tag validation bits in a cache memory, tag modified/dirty bit in a copy-back/write-back cache memory and the like. The implemented tag memory further stores the address field and attributed field using memory units with the structure shown in FIG. 2 a.

Next, in step S 400 , the implemented tag memory is simulated to check for the conditions determined in step S 200 . If not, the procedure returns to step S 200 .

Step S 500 is executed if the implemented tag memory meets the conditions determined in step S 200 . In Step S 500 , peripheral elements in the cache memory are implemented according to the specification for the desired cache memory. For example, a D-type Flip-Flop and a driving buffer are implemented in the cache memory to provide a flush control signal to the tag memory for flush. In Step S 600 , the peripheral elements are then modified to match the tag memory, such that the peripheral elements meet the desired specification. For example, a D-type Flip-Flop and a driving buffer are modified repetitively to provide a flush control signal with the requisite time cycle. Finally, step S 700 is executed if the peripheral elements meet the desired specification. In step S 700 , integration of the tag memory and the peripheral elements in the cache memory is simulated to ensure that integration of the tag memory and the peripheral elements meets the desired specification. If not, this procedure returns to step S 200 . If so, the cache memory is complete with the desired specification.

While the invention has been described by way of example and in terms of the preferred embodiments, it is to be understood that the invention is not limited to the disclosed embodiments. To the contrary, it is intended to cover various modifications and similar arrangements (as would be apparent to those skilled in the art). Therefore, the scope of the appended claims should be accorded the broadest interpretation so as to encompass all such modifications and similar arrangements.