Multiple data path memories and systems

Abstract

The invention describes and provides multiple data path memories and systems utilizing such memories. Enhanced data throughput and bandwidth, while substantially simultaneously providing improved bus utilization, are some of the benefits. In peer-to-peer connected systems, multiple bank/access block/sector/sub-array with random data throughput can also be realized. A memory including a plurality of independently accessible memory banks, a READ BUS for selectively reading to a selected on of the memory banks, and a WRITE BUS independent of the READ BUS for selectively writing to a selected one of the memory banks, is described.

Description

BACKGROUND OF THE INVENTION

[0002] Field of Invention

[0003] The present invention relates in general to semiconductor memories. Those memories can be standalone, embedded, or stacked as multiple chips (homogeneous and heterogeneous—namely, in a memory stack, all of them can be of one kind; or, they can be different in function, utility, form factor as well as to technology).

BACKGROUND OF INVENTION

[0004] Semiconductor memories include dynamic random access memories (DRAM's) static random access memories (SRAM's), flash, ferroelectric, magnetic, plastic, MIM (metal-insulator-metal) RAMs, calcogenide and plastic memories, and others. The prevalent monolithic and embedded memory architecture, is multi-bank. The input/output data architecture includes double data rate (DDR), quad data rate (QDR) zero bus turnaround (ZBT SRAM's) and others. Memory storage cell architecture in nonvolatile memories, specifically flash, could be NOR or NAND (others are possible as well). As an example a conceptual block diagram of a commercially available Infineon RLDRAM (reduced latency DRAM) is shown in FIG. 1.

[0005] With the advent of diverse applications for both volatile and nonvolatile memories, a need has arisen to further enhance bandwidth and bus utilization. In peer-to peer system, as well as bus oriented systems, a need has arisen to access multiple banks/simultaneously blocks/sectors/sub-arrays, for maximum data throughput. Spin-wheel SDRAM implementations are another example of systems to address these needs. One specific application where data moves in packets (both fixed length and variable length packets) is, in communications. ATM, IP, SONET, Ethernet are some of the protocols used in communicating the information as digital data. The memory buffers used in such communications have to be flexible enough in configurability to accommodate varying length packets, with maximum efficiency (and not waste memory space). Because of the criticality of packet transmission and reception in a predetermined QOS (quality of service) deadline, each packet must be distributed across multiple banks in multiple IC's, in both INGRESS (WRITE) and EGRESS (READ) of those packets. Multiple data path memories can handle these time critical decisions without ‘loss of packets,’ which is not possible with current memory, architectures.

SUMMARY OF INVENTION

[0006] The invention provides a memory including a plurality independently accessible memory banks (or blocks, or sectors, or sub-arrays as they are called in the semiconductor industry to denote memory arrays) organized as rows x columns), a read bus for selectively reading to selected one of the memory banks (or blocks, or sectors, or sub-arrays), and a write bus independent of the read bus for selectively writing to a selected one of the memory banks (or block, or sectors, or sub-arrays). The word bank “bank” is used here synonymously with block, sector, sub-array etc.

[0007] Among the numerous advantages of the various embodiments of the principles of the present invention, is to independently read and write to the banks in a multiple bank memory system. One embodiment utilizes independent read and write buses to accomplish this task. For example, while a read is being performed to one bank in the system (or IC) using the read bus, a write to another bank in the same (or different) system (or IC) can be independently and concurrently taking place using the write bus. In another embodiment, specifically where the ‘read’ and ‘write’ bus have to be ‘reset’ prior to the next clock (‘reset’ is a term that is used synonymously with ‘precharge’ or ‘set for next operation’ or ‘next active command on that line’), one of the read and write buses can be pre-charged while the other is used to access data in an (active) addressed memory bank. In further embodiments, the independent read and write buses are associated with independent read and write data ports in the input/output space supporting independent data exchanges with other external components.

BRIEF DESCRIPTION OF DRAWINGS

[0008] For a more complete understanding of the present invention, and the advantages thereof, reference is now made to the following description in conjunction with the accompanying figures:

[0009] FIG. 1 is a block diagram of an exemplary Infineon's DDR reduced latency DRAM (HYB18RL 25632AC/AYB18RL25616AC).

[0010] FIG. 2 is a conceptual diagram of a typical packet used in ATM, IP, SONET and similar protocols using packet data exchanges.

[0011] FIG. 3 is a conceptual system diagram for a payload (packet buffer) memory.

[0012] FIG. 4 is a block diagram of a new memory with separate read and write paths according to the principles of the present invention. Memory core (array) can be of any kind —DRAM, SRAM, flash, ferroelectric, magnetic, etc.

[0013] FIG. 5 is a timing diagram illustrating the timing of a Read command.

[0014] FIG. 6 is a timing diagram illustrating the timing for a READ COMMAND and/or WRITE command to two different banks.

[0015] FIG. 7 is a block diagram of a Micron 288 Mbit separe I/O RLDRAM II (MT 49H 16M18C, MT49H32M9C) and

[0016] FIG. 8 is a block diagram of a 288 Mbit separate I/O RLDRAM II according to the principles of the present invention (the memory array/core can be DRAM, SRAM, flash, ferroelectric, magnetic, etc.).

DETAILED DESCRIPTION OF THE INVENTION

[0017] The principles of the present invention and their advantages are best understood by referring to the illustrated embodiments depicted through FIGS. 1-8 of the drawings, in which like numbers designate like parts. FIGS. 2 and 3 describe the basic operation of a communication packet buffer memory as it is utilized today in communication systems (base stations, routers, hubs, switch fabrics). References to prior art, are cited along with this patent.

[0018] The invention described here, as illustrated in FIG. 4 has a separate READ BUS and separate WRITE BUS coupled to the sense amplifiers of each memory bank (array unit). This separation of READ and WRITE data path out of the sense amplifier significantly boosts the peak bandwidth as well as the random cycle time at the IC level itself. One method of implementing these buses to sense amplifiers has been illustrated in U.S. Pat. No. 6,396,764.

[0019] FIGS. 5 and 6 are timing diagrams comparing the enhancements obtained with invention. One set of row address buffers and column address buffers is preferred; however, separate row and column address buffers for READ and Write addresses are not excluded in this invention. The former eliminates the “dead time” for bank access. In today's commercially available RLDRAMs™, (HYB18RL25632AC, MT49H8M32), have 8 banks each in monolithic 256/288 MbDRAM, FIG. 7 illustrates this as an example (Micron's RLDRAM). These 8 banks are sequentially cycled for READ or WRITE, one bank at a system, at a system clock frequency (200-400 MHZ). Typically, a bank is available only every 8th cycle—in other words, if Bank 0 is accessed in cycle 1 (for READ or WRITE), the next access to the same Bank 0 has to wait till cycle 9 (assuming one wants access to other banks in the intermediate cycles—the most likely scenario, in systems operation). Sequentially, READING or WRITING is allowed, for data bursts, once a bank is OPEN (active)—however, such “burst” of data is in a predetermined address sequence/and hence, not random address). However, there are timing restrictions for READ and WRITE sequences to different banks within those 8 cycles. These restrictions are due to common input/output pins, data bus unavailability due to double data rate streaming in a burst mode, ability to communicate precisely on a bus with timing preambles, as they are known in data sheets and timing diagrams. Separate READ and WRITE buses remove most of the above impediments and restrictions, for maximum efficiency. Separate Input/Output data pins further enhance data bandwidth. For example, as shown in FIG. 4, with independent READ and WRITE data paths, data can be READ OUT or WRITTEN IN even with a burst mode DDR. This invention does not restrict further improvements like quad data rate (QDR). Preferably, during a burst length of 4 or greater, the ‘active’ (current) bank precharge can be accomplished in the background, thus improving random latency for a new row, even further. Separate I/O pins, while still maintaining common data path to the sense amplifier, as shown in Micron's RLDRAM, is still restrictive.

[0020] FIG. 8 shows another embodiment of this invention. Each bank (memory array unit) has a separate data path next to its sense amplifier. This embodiment can further enhance data throughput.

[0021] It is also obvious that the memory array (core) for data storage can be any volatile and nonvolatile memory. The inventions described here are accommodative of any of those memories.

[0022] While some particular embodiments of the invention have been shown and described, changes and modifications may be made therein without departing from the invention in its broader aspects. Therefore, the aim in the appended claims to cover all such changes and modifications as fall within the true spirit and scope of the invention.

Claims

1. A memory system comprising at least one memory integrated circuit having a plurality of independently accessible memory banks a read bus for selectively reading to a selected one of the memory banks a write bus independent of the read bus for selectively writing to a selected one of the memory banks.

2. A memory integrated circuit comprising separate read bus and write bus, in conjunction with separate input and output pins.

3. A memory integrated circuit comprising separate read and write bus with common input/output pins.

4. A memory integrated circuit of claim 1 containing memory cells, requiring periodic data refresh.

5. A memory integrated circuit of claim 1 containing static memory cells not requiring periodic data refresh.

6. A memory integrated circuit of claim 1 containing nonvolatile memory cells.

7. A memory integrated circuit of claim 1 containing a combination of dynamic, static and nonvolatile memory cells.

8. A SIP (System In Package) containing at least one integrated circuit covering any of the embodiments of claim 1.

9. A SOC (System On Chip) containing at least one embodiment of claim 1.

10. A memory integrated circuit containing at least one embodiment of claim 1, that can operate in a shared common bus environment.

11. A memory integrated circuit containing at least one embodiment of claim 1, that can operate in a peer-to-peer environment.

12. A memory integrated circuit containing at lease one embodiment of claim 1 with rail-to-rail electric signaling.

13. A memory integrated circuit containing at lease one embodiment of claim 1 with single ended electrical signaling.

14. A memory integrated circuit containing at least one embodiment of claim 1 that can operate synchronously.

15. A memory integrated circuit containing at lease one embodiment of claim 1 that can operate asynchronously.