Skew free control of a multi-block SRAM

1. Field of the Invention

This invention relates to the control of multi-block semiconductor SRAM, and more particularly to the generation of memory control pulses.

2. Description of the Related Art

Recent designs of Megabit semiconductor SRAM (static random access memory) are generally based on a multi-block architecture as shown in FIG. 1 . The multi-block architecture design results in power saving and memory speed improvement. The central control generates global signals which are distributed to each memory block where they are used directly or are decoded to produce required local control signals. The larger the size of the SRAM, the longer the distribution paths for the global control signals. The global control pulse signals travel to each memory block and have different wire loading and fan-out loading causing a variation in delay at each memory block. Memory design requires a specific relationship between the leading and trailing edges of control pulse signals. The differences in loading and fan-out result in variation in skew of these control pulses. In high speed memories, these variations in loading and fan-out are significant and limit the speed of memory operation by requiring sufficient timing margins to accommodate worst case skews.

FIG. 2 and FIG. 3 illustrate the effect of skew on the timing relationship between memory control pulses for Read and Write operations respectively. The correct timing relationship between the memory control pulses exists at the output of the Central Control ( FIG. 2A ). These are globally distributed to each memory control block. These signal pulses have varied delays as they propagate through the system which affects the timing relationships between the pulses. In the extreme case, the resultant skew may be large enough to cause memory malfunction. This is illustrated for the Read operation in FIG. 2 . Two cases are shown. In case 1 ( FIG. 2C ) the Word-Line pulse has a greater delay than SA enable. The limit at which the timing relationship is still maintained is shown occurring at Block M ( FIG. 2B ). In subsequent memory blocks the skew exceeds the required timing criteria resulting in operational failure, as illustrated for Block N. In case 2 ( FIG. 2D and FIG. 2E ) an analogous situation occurs affecting the trailing edge timing relationships. FIG. 3 illustrates similar situations for the memory Write operation.

Four known U.S. patents deal with multi-block systems involving either global interconnection or local generation of control signals.

U.S. Pat. No. 5,837,557 (Fulford, Jr. et al) describes a semiconductor structure consisting of multiple circuit blocks which are globally interconnected by means of a blanket metal layer. The structure described in the patent allows a reduction of overall circuit size, enhanced circuit quality and improvements in fabrication time and costs.

U.S. Pat. No. 5,581,126 (Moench) deals with the layout of interconnect lines within integrated circuits. It is the object of the invention to eliminate data pattern sensitivity and unwanted differential signals in SRAM arrays due to lateral capacitive coupling from adjacent lines. This is accomplished by using a crossover structure to reduce the effect of parasitic capacitance.

U.S. Pat. No. 5,243,703 (Farmwald et al) describes a method for generating a synchronized clock pulse at each device block in a computer memory system consisting of multiple blocks. This is done with a single globally generated clock pulse that is transmitted to each of the blocks via a looped path. Each block receives one input from the forward half of the loop and another input from the return half of the loop. This allows each block to generate a clock pulse with synchronized timing. The patent does not address the problem of control pulse skew as described by the present invention.

U.S. Pat. No. 4,926,384 (Roy) deals with SRAM write cycle recovery timing. The patent describes an SRAM composed of multiple separate memory blocks where each block has its own local bit line equalization circuitry. This removes the process of equalizing the bit lines after a write cycle from the critical timing path for accessing the memory. The patent also provides means for automatically equalizing the common data out lines at the end of each memory access cycle which removes the process of equalizing the common data out lines from the critical timing path for accessing the memory.

It is the object of this invention to reduce the timing skew between memory control pulses in multi-block semiconductor memory systems.

Another object of the present invention is to use the reduced timing skews between memory control pulses to achieve faster memory cycle times.

Another object of the present invention is to use the reduced timing skews between memory control pulses to increase SRAM operational reliability.

A further object of the present invention is to use the reduced timing skews between memory control pulses to increase the SRAM chip yield.

These objects have been achieved by generating memory control pulses locally in each memory block. A single global pulse, distributed to each memory block, initiates the generation of local memory control pulses.

FIG. 1 shows a block diagram of a conventional multi-block SRAM with centrally generated and globally distributed memory control pulses.

FIG. 2 ( FIG. 2A , FIG. 2B , FIG. 2C , FIG. 2D , FIG. 2E ) shows the effect of skew on timing relationship during a SRAM Read operation

FIG. 3 ( FIG. 3A , FIG. 3B , FIG. 3C , FIG. 3D , FIG. 3E ) shows the effect of skew on timing relationship during a SRAM Write operation

FIG. 4 shows a high level block diagram of multi-block SRAM using a single global pulse with locally generated memory control pulses.

FIG. 5 shows a high level block diagram of a local memory control pulse generator.

Referring now to FIG. 4 , it shows a block diagram illustrating the principal components, according to the preferred embodiment of the invention, consisting of central control and a multiplicity of SRAM memory blocks. The Central Control 21 generates a single Global Pulse which is distributed to each of the memory blocks via Global Pulse Line 20 . Each of the SRAM memory blocks 22 , 23 and 24 in the system are identical. Each memory block contains an identical local signal generator which responds to the single global pulse and generates the required memory control pulses. Thus, SRAM memory block 23 (M) contains the local signal generator (Loc.Sig.Gen.) 25 which generates the four memory control pulses which are required for the various memory operations. These are: Word Line pulse (WL), Bit-line Pre-charge pulse (BL Prech.), Sense Amplifier Enable (SA Enab.) and Write Enable Pass Gate (WEPG).

FIG. 5 shows a block diagram of a local signal generator. A similar local signal generator is contained at each memory block. The function of the local signal generator is to generate all the needed local pulse signals for the read and write memory cycles with the correct relationship between the leading and trailing edges of the memory control pulses. These relationships are controlled by means of delay elements as shown in FIG. 5 . For example, the circuit comprised of Gate 31 , Delay Element 32 , and Gate 33 generates the SA Enable control pulse with a leading edge delayed from the leading edge of the Global Pulse and a trailing edge aligned with the trailing edge of the Global Pulse. All local control pulses are enabled by the local Block Enable level. Similarly, in FIG. 5 , the other control pulses are generated with the required leading and trailing relationships by means of appropriate delay elements. This technique for generating pulse relationships is well known to those skilled in the art.

The advantages of the present invention are:

it eliminates the skew between memory control pulses that occurs in systems with centrally generated and globally distributed memory control pulses; Improvement in memory cycle time by reducing the required timing margins; Improved operational reliability by maintaining proper memory control pulse relationships without increased timing margins; chip manufacturing process variations have reduced impact on skew thereby improving chip yield.