Field of the Invention: The present invention relates generally to a method and apparatus for capturing high-bandwidth commands and addresses. More particularly, the present invention relates to an address FIFO (first-in first-out) procedure and apparatus for use with a data storage system such as dynamic random access memory (“DRAM”). A FIFO circuit buffers incoming memory address commands until corresponding data arrives to smooth data transfer to memory. The invention also relates to a FIFO buffer system that maintains the write pointer at least one buffer ahead of the read pointer to enable loading of buffers while others are being unloaded.
State of the Art: Faster and smaller circuits are the focus of much advancement in semiconductor technology. To address the need for faster and smaller circuits, a group of integrated circuits can be on a common bus. In this configuration, each integrated circuit operates in a coordinated manner with the other integrated circuits on the bus to share data which is transmitted at a high speed. An example of such a high-speed data system is described in U.S. Pat. No. 5,917,760 to Millar (Jun. 29, 1999), the disclosure of which is incorporated herein by reference. Millar describes a high-speed data system using a common bus and a memory subsystem commonly known as SyncLink dynamic random access memory (“SLDRAM”). By providing an appropriate number of memory devices and an efficient control system as used in SLDRAM, very high-speed data transmissions can be achieved. However, faster systems, such as SLDRAM, are now reaching transfer speeds where the memory circuits cannot process the data as fast as the common bus can supply it. More specifically, as data storage address commands arrive at a memory system, if the corresponding data has not arrived yet, the address commands must be stored in a data pipeline until the data arrives. This address command backlog prevents use of the data pipeline by other processes, including transmitting other data, until the data corresponding to the address command arrives. This inconsistency in address command and data arrival times can result in increased data errors and lost data, but most often slows the system by creating a “bottleneck” of address commands.
A pipeline may be divided into any number of stages during which portions of commands are processed and executed. However, in a case of a memory device, such as SLDRAM, the series of processes typically includes: 1) an input process of address data; 2) a decoding process of the address data; 3) a reading process of data from a cell; 4) a transfer process of the data to an output circuit; and 5) an output process of the data. An example of a pipeline system used with DRAM is provided in U.S. Pat. No. 5,978,884 to Yamaguchi et al. (Nov. 2, 1999), the disclosure of which is incorporated herein by reference.
One solution which has been used in memory pipeline systems to correct for the bottleneck problem is to repeat much of the logic circuits within each logic pipeline to accommodate multiple simultaneous commands. However, adding more redundant logic circuits to a system is counter to the desire to make the overall system smaller.
Many telecommunication devices include first-in first-out (FIFO) circuits which temporarily buffer data arriving at a bandwidth higher than the bandwidth of the receiving system. A FIFO circuit stores incoming data, in the order it arrives, in temporary buffers which are then sequentially read out and used by the telecommunication subsystem for which they were intended. The FIFO circuit can store the mass of data that arrives before it can be processed in a temporary storage and read it at a manageable speed. For example, a high-bandwidth data signal can be received at any speed by a telecommunication device, stored in the FIFO buffer, and read out at the processing speed of the device.
One example of a FIFO circuit used in a telecommunication system includes U.S. Pat. No. 4,507,760 to Fraser (Mar. 26, 1985). Fraser discloses random access memory (RAM) organized to act as FIFO memory and a control circuit to implement queue management for incoming/outgoing data in a digital communication system. A read pointer addresses the execution in the RAM from where a word may be read. A write pointer addresses the location in the RAM where a word may be entered. A “last” pointer addresses the location in the RAM where the last word of a complete message is stored.
Another example of a FIFO circuit used in a data communications system is described in U.S. Pat. No. 5,519,701 to Colmant et al. (May 21, 1996). Colmant et al. disclose a system to manage storage of data in FIFO circuits as data is transferred, in either direction, between the host bus and the network. By a queue manager allocating the queues, which have the most activity, the queue manager can improve the speed of the transferring data while reducing the amount of bandwidth that would otherwise be required.
It is desirable to have a memory system that can handle the increased speed demands made by faster circuits, preserve data which arrives faster than the memory circuits can handle it, and do so without the redundancy of circuits required by existing memory systems.
The present invention addresses the problem of the bottleneck created in high bandwidth to smaller bandwidth systems through the use of a first-in first-out (FIFO) buffer system. The FIFO buffer system of an embodiment of the invention comprises a sequential series or plurality of FIFO buffers associated with read and write counters or pointers to indicate the next FIFO buffer in the sequential series from which data should be read, or to which data should be written. Read and write address decoders are coupled between the FIFO buffers and the read and write counters, respectively, to decode the pointer indicators to a particular FIFO buffer indicator. Of particular interest within the FIFO buffer system is the way in which the read counter operates. The read counter tracks both the current and previous setting for the read counter, yet indicates through the decoder to the FIFO buffers the previous setting as the particular buffer to which the read counter is pointing. A result of this form of operation is that the first read latch signal sent to the read counter is ignored so far as the read counter's indicator is concerned. Following a reset signal, the read counter points to the first of the series of FIFO buffers even after receiving the first read latch signal. Subsequently, the pointer increments one buffer in the sequential series of FIFO buffers for each read latch signal received. Another result of this form of operation is that through the control logic associated with the FIFO buffer system, the write counter always points at least one FIFO buffer ahead of the FIFO buffer to which the read counter is pointing. This allows the FIFO system to read buffers while writing to other buffers.
In particular use with a dynamic random access memory (DRAM) device, the FIFO buffer system stores address commands until corresponding data to be stored arrives at the DRAM. This frees up the DRAM pipelines for use with transferring data rather than storing address commands for the data.
In one embodiment of the FIFO buffer system within a DRAM device, the read counter comprises a plurality of registers to track the current and previous register settings. A two to four decoder is used to decode a two digit binary code indicating the previous state into a signal to activate one of four decoder outputs corresponding to related FIFO buffers. For each decoder output activated, a column address buffer and a row address buffer are activated.
In another embodiment of the FIFO buffer system, the read counter comprises a linear feedback shift register to track both the current and previous register settings and indicate the previous register setting as the FIFO buffer to which the read counter is pointing.
The nature of the present invention as well as other embodiments of the present invention may be more clearly understood by reference to the following detailed description of the invention, to the appended claims, and to several drawings herein, wherein:
One aspect of the FIFO buffer of the present invention, distinct from counters used in telecommunication systems, is that the read address decoder circuit takes, as the pointer of the buffer address to be read, the previous counter setting rather than the current counter setting. As a result, when the access signal fires to determine whether significant data is contained in the FIFO buffers, the read pointer, which was at address 0 prior to the firing, is still pointing at address 0 after firing. In other words, the first access signal is ignored as far as the pointer is concerned. By using the previous counter setting as the pointer indicator to read from, the first buffer register can contain significant data. Additionally, by ignoring the first access signal, the write address pointer is always at least one position ahead of the read address pointer. A result of maintaining the write address pointer at least one position ahead of the read address pointer is that other registers can be loaded while a register is being read. Without that space, a processor was required to wait until a write process was completed to begin a read process. With the space, the read latency is reduced. Thus, in the present embodiment, the read counter 6 outputs both the current and previous counter setting positions of the FIFO buffer registers.
An alternative to having the read pointer not advance on the first read operation is to set the read pointer reset value as the last address and the write pointer reset value as the first address. This solution, however, may cause problems with read pointer decoding latency and is therefore not preferred. Two reasons for having the read pointer remain at the first address after the first read access are: 1) consistent logic implementation lending itself to pipeline architectures; and 2) masking of read pointer decode latency on subsequent accesses.
As will be understood by one of ordinary skill in the art, there are numerous memory bank address commands which need to be transferred in addition to the Read and Write addresses. Examples of memory bank address commands include Bank Address commands, Restore commands, Auto Precharge commands and Burst commands. Each of the memory bank address command bits is transferred to and stored in the FIFO buffer along with the Read and Write addresses to be buffered until the memory is ready to receive it. The necessary memory bank address command bits and methods and apparatus for enabling their production are well known in the art.
In operation, after the reset signal fires through reset terminal or input 22, each of the registers 18, 20, 26, 28 and 30 in both the write and read counters 4 and 6, also called pointers, are targeted at address 0. Being targeted at address 0 means they are currently set to read from and write to both the column and row FIFO buffers 046 and 48. A write latch signal through write latch terminal or input 24 and a read latch signal through read latch terminal or input 32 are each respectively used to toggle the write counter (or write address pointer) 4 and read counter (or read address pointer) 6. At some time after the first address is latched into FIFO buffers 046 and 48, the read latch signal may be asserted to read out the oldest data in the FIFO buffer circuit, for this case, namely the first address latched into FIFO buffers 046 and 48. On the first read latch signal, the first and second registers 26 and 28 of the read counter 6 increment by one counter setting to point at the FIFO buffers 150 and 52. However, the serial combination of the outputs from the second and third registers 28 and 30, which indicates the previous counter setting rather than the current counter setting, still points at the FIFO buffers 046 and 48. Because the read address decoder 10 takes as its input the output from the third register 30 and combines it in serial order with the output from the second register 28, the read counter/decoder combination has, in essence, ignored the first read latch signal. Thus, even after the first read latch signal, the read counter 6 is still pointing to the first FIFO buffers 046 and 48. Each successive firing of the read latch signal through read latch terminal or input 32 will move the read address pointer sequentially ahead one FIFO buffer register. This automatically causes a minimum of one buffer position offset between the read and write pointers. The result of this operation is, while a register is being read, other registers can be loaded since the write pointer is at least one position ahead of the read pointer. There is an assumption, however, that to maintain this relationship, every read pointer change requires at least one preceding write pointer change. It will be obvious to one of skill in the art how to program the logic controlling this circuit to maintain this relationship.
The write and read address decoders 8 and 10 are conventional 2 to 4 decoders, meaning that they take a binary input of two bits and translate it into a signal on one of four outputs corresponding to the value of the two bit binary input. Write signals 40 output from the write address decoder 8 are Write_<0>, Write_<1>, Write_<2> and Write_<3>. Each of these signals as well as its inverse, created by inverters 44, is fed to two of the FIFO buffers 12 corresponding to the number within the brackets < > following the signal type. For example, the Write_<0> and Write <0> signals are fed to each of the FIFO buffers 046 and 48. Similarly, the Read signals 42 output from the read address decoder 10, Read_<0>, Read_<1>, Read_<2> and Read_<3>, along with their inverse, are each fed to two of the FIFO buffers 12 corresponding to the number within the brackets < > following the signal type. Thus, Read_<1> and Read_<1> are both fed to each of the FIFO buffers 150 and 52.
Each of the addresses stored in the FIFO buffers are read out on a first-in first-out basis through column and row address output circuits 14 and 16 which translate and delay the addresses as required by the DRAM in which the circuit 2 of the memory bank control logic circuit is used. It should be understood that, although the preferred embodiment is shown with only four FIFO buffers, it is contemplated that any number of buffers may be used according to the principles taught herein by simply increasing the number of registers in both the write and read counters and appropriately increasing the number of registers in the FIFO buffers.
As shown in
As will be clear to one of ordinary skill in the art, the FIFO buffer system shown and described herein, though depicted as an address buffer for DRAM, is not limited to application in DRAM. One of ordinary skill will understand how to apply particular embodiments of the FIFO buffer system to other systems where a FIFO buffer system may be useful, such as telecommunications systems.
Although the present invention has been shown and described with reference to particular preferred embodiments, various additions, deletions and modifications that are obvious to a person skilled in the art to which the invention pertains, even if not shown or specifically described herein, are deemed to lie within the scope of the invention as encompassed by the following claims.
This application is a continuation of U.S. patent application Ser. No. 11/496,302, filed Jul. 31, 2006, scheduled to issue as U.S. Pat. No. 7,698,499 on Apr. 13, 2010, which is a continuation of U.S. patent application Ser. No. 10/644,324, filed Aug. 20, 2003, now U.S. Pat. No. 7,085,880, issued Aug. 1, 2006, which is a continuation of U.S. patent application Ser. No. 09/541,732, filed Apr. 3, 2000, now U.S. Pat. No. 6,732,223, issued May 4, 2004, the disclosure of which is hereby incorporated herein by this reference in its entirety.
Number | Date | Country | |
---|---|---|---|
Parent | 11496302 | Jul 2006 | US |
Child | 12758598 | US | |
Parent | 10644324 | Aug 2003 | US |
Child | 11496302 | US | |
Parent | 09541732 | Apr 2000 | US |
Child | 10644324 | US |