Pipelined semiconductor memories and systems

Abstract

The invention describes and provides pipelining of addresses to memory products. Addresses are pipelined to multibank memories on both rising and falling edges of a clock. Global Address Supervisor pipelines these addresses optimally without causing bank or block or subarray operational conflicts. Enhanced data through put and bandwidth, as well as substantially improved bus utilization (simultaneously), can be realized. In peer-to-peer connected systems, significant random data access throughput can be obtained.

Description

FIELD OF INVENTION

[0002] The present invention relates in general to electronic systems comprising semiconductor integrated circuits. It relates in particular to pipelined memories in standalone (discrete) as well as embedded (system-on-chip, system-in-package) implementations.

BACKGROUND OF THE INVENTION

[0003] Peak data bandwidth, average data bandwidth, fast bus turnaround, maximum bus utilization and efficiency, low power, nonvolatility—all at an affordable cost—are key requirements for semiconductor components. Specifically, for semiconductor memories, there are additional requirements as well. For example, balanced read/write operational efficiency in communication systems, is necessary. In some systems dominated by ‘page’ architectures (DRAM, Flash are some examples), multiple open pages improves system efficiency. Since memory integrated circuits are used in large numbers in electronic systems, their ability to function efficiently in bus architectures, as well as peak-to-peak architectures is desirable.

[0004] Most memories, at the core, are two-dimensional arrays of rows and columns. DRAMS, Flash, SRAMs, EEPROMS, Ferroelectric memories, Magnetic RAMS, nanotube RAM's (carbon nanotube is one example), molecular memories, phase change memories and organic memories etc. Each of these memories serve a particular application satisfying the requirements of that particular application. Although, all these memories are Read and Write memories, each application requires focus on a particular parameter. ‘Open page’ applications (Personal Computers, Servers for example) require fast data with a given page (or predetermined set/sets of columns)—Rambus™, DDR (double data rate), QDR (quad data rate), FCRAM™, RLDRAM™ (reduced latency) are serving those requirements. ZBTSRAM™ serves some specific needs in wired communication systems. Pseudo static RAM'S, nonvolatile SRAM'S, MIM (metal-insulator-metal) RAM'S are finding acceptance in portable electronic systems like cellular phones.

SUMMARY OF THE INVENTION

[0005] All of the above memories without an exception, desire fast data throughput at low cost and low power. Although data pipelining and prefetching from the memory core have been described in prior art, address pipelining in general, bank/block/sector/row/subarray and address pipelining in particular, have not been attempted. It is one embodiment of the invention to pipeline all addresses at the same rate as data, independent of the address function (row, column, bank, block, sector). It is another embodiment of the invention to pipeline those addresses on both the rising and falling edges of SCLK (System clock) or a primary clock for that particular system or IC. It is yet another purpose and embodiment of the invention to provide a global command and control supervisory circuitry for each monolithic integrated circuit that optimizes the efficiency of a multi bank/block/sector IC. The word “bank” is used here, synonymously with block, sector, subarray etc. It is also another embodiment to pipeline addresses at a rate faster than data, or, even slower than data in an asynchronous manner.

[0006] Although preferred embodiments are described in this invention, the implementation and extension of the principles of the invention are not limited. For those skilled in the art, the principles described in this invention will be obvious. The principles of the present invention are embodied in memory system architectures and operating methods utilizing multiple banks (blocks, sectors, subarrays) and independent row/column address decoding pipeline. A memory is disclosed where a plurality of independently addressable banks for storing data can function with a decoding pipeline of n-stages where n is greater than 1, and at least 2. The “unit” is one system or primary clock (SCLK, CLK) period. Row and column addresses can be continuously pipelined from the address input parts. Global address supervisory circuitry allows a sequence of addresses to be continuously received, and, properly implemented, without conflict at any stage.

BRIEF DESCRIPTION OF THE DRAWINGS

[0007] FIG. 1 is a functional block diagram of a 32 Meg×16 double data rate synchronous RAM (memory) incorporating some of the inventions like global supervisory control described in this patent.

[0008] FIG. 2 is a functional block diagram of a 32 Meg×16 memory with additional embodiments of the invention where the global supervisory control specifically focuses on the bank and row address path.

[0009] FIG. 3 is a functional block diagram of a memory system/IC where an SRAM buffer/register between the memory core (can be any memory, DRAM core is shown as an example only) and peripheral circuitry (address, command, control and clocks).

[0010] FIG. 4 is a functional block diagram of a memory system/IC where the READ and WRITE data paths are separately gated to the core memory.

DETAILED DESCRIPTION OF THE INVENTION

[0011] The principles of the present invention and their advantages are best understood by referring to the illustrated embodiments depicted in FIGS. 1-4 of the drawings, in which like numbers designate like parts. The inventions described below apply to any memory, namely, DRAM, SRAM, EPROM, EEPROM, Flash, Mag RAM, FeRAM, PCRAM, plastic RAM, CNTRAM, Molecular RAM etc. The inventions apply to both non-multiplexed address as well as multiplexed-address integrated circuits. The inventions described below apply to what is known in the industry as “row” chain—namely, selecting block/bank/sector/subarray/row in any IC (integrated circuit) or system (consisting of several IC's) or SOC (System On Chip). The inventions apply to single-ended or rail-to-rail address/data/clock signals. The inventions apply to the “column” chain as well. The inventions apply to “data bursts”, “prefetch schemes”, “page schemes” and similar architectures known in the industry. The effective bandwidth of a device can be enhanced 2 to 4× or even more, by using these inventions, with minimal cost.

[0012] The time line sequence of events in a monolithic commercial multi-bank synchronous DRAM (generally speaking)—a 512 Mb DRAM, organized as 32 Mb×16, for example is described below (a clock, a command, a control are given, in addition to the addresses, to access one or more sets of memory locations). For simplicity's sake, we shall describe these in ‘clock units’ (1 unit is one master clock period). Assume all banks have been precharged before the memory access cycle begins (it is assumed and understood that MODE registers, defining specific device operational choices are set appropriately):

[0013] a) Addresses are received, appropriately level shifted/converted after detection (on chip)

[0014] b) Bank selection is made

[0015] c) Row address is streered to the selected bank

[0016] d) Row address is decoded (to select 1 of n rows) in the specific bank (assumes proper subarray is selected, if necessary)

[0017] e) Word line is activated

[0018] f) Signal from all storage cells (memory cells) in the related row, are detected and amplified by the sense amplifier (for a read operation in a dynamic cell, appropriate digital level is written instead of RESTORE)

[0019] g) Column start address is selected, data prefetched in the ‘burst order’ defined (activation of column address select can be posted earlier; ‘posted CAS’ as is well known in the industry)

[0020] h) If the command is READ, appropriate data is sent to output buffers (depending on data fetch schemes, DDR or QDR is employed

[0021] i) If the command is WRITE, appropriate data is received from the input buffers (level shifted and amplified as needed) is coupled to appropriate lines of columns at the sense amplifiers

[0022] j) RESTORE is performed if a memory core requires that function, automatically, as a part of the READ

[0023] k) Page is left open for further READ/WRITE operations into the same PAGE (usually one Page is equal to one row of data). As is well known in the industry, nonvolatile memory IC's also used PAGE driven architecture, by employing an SRAM buffer.

[0024] In the case of RLDRAM (reduced latency DRAM), the banks are cycled in sequence and the PAGE in the previously accessed bank is closed in each bank as the cycling starts to a new bank access.

[0025] Precharge is not required for 6TSRAM's, Flash nonvolatile memories and similar memory cores. Precharge is required PSRAM's (pseudo static RAM's) as is know in the industry. The above ‘time sequence’ varies for memory type, namely, DRAM, SRAM, flash etc.

[0026] The critical paths of ‘row access’ and ‘column access’ dictate the ‘access time’ of each memory. In today's popular memory products (DRAM's, SRAM's, flash), although the memory may be in multiple banks in the same monolithic chip, the peripheral circuitry (internal clocks, data paths to the external media) is common. This limitation, significantly, impedes access speed. For example, while one ‘row’ (wordline) is ON, the next row address for access can be waiting in line without turning ON the wordline. As illustrated in FIG. 1, the invention highlights what has not been shown in prior art. Firstly, addresses are toggled on both raising and falling edges of the CLK. Effectively, this doubles the rate (speed) at which addresses can be funneled into the IC, compared to prior art. Secondly, all incoming addresses (after level shifting and amplification, as necessary) directly go to the Global Address Supervisor (global address/command/control register and scheduler). Such an architectural item is not know in the industry, nor implemented in any of the IC's (DRAM, SRAM, flash, NVSRAM). The Global Address supervisor, in this invention, performs some of the following tasks:

[0027] 1) it steers the address to the designated bank (if a DRAM or SRAM) or block/sector (if Flash memory) or similar memory array unit as used by memory IC's.

[0028] 2) it has the ability to map, if required, addresses, or sequence the addresses appropriately, to avoid bank/block/sector/subarray/row conflicts

[0029] 3) if a BANK is busy, it has the ability to hold the addresses in a STAGING area temporarily, and release that address to that bank at the appropriate time without causing device malfunction. This increases bus utilization time.

[0030] The Global Address Supervisor is shown in FIGS. 1 and 2. Whereas FIG. 1 highlights the invention for a multiplexed address DRAM, FIG. 2 shows the implementation of this invention to any memory (not limited to DRAM). The staging area for the addresses—row, column or whatever—can be implemented either with SRAM like latches (set/release latches) or similar digital logic circuitry well known in the industry. Such isolation latches can also be used in various stages of clocked circuitry, where needed in the row path as well as column path.

[0031] It should be obvious that the peak as well as average bandwidth of the memory IC or system can be doubled, with minimal cost. By employing separate READ and WRITE access port, the bandwidth can be increased even further. By employing FIFO'S (FIRST IN, FIRST OUT) in the WRITE path, one can further enhance performance in specific application like packet buffering (where the INCOMING PACKET QUEUE is well defined, where as the OUTGOING PACKET QUEUE is mostly random).

[0032] Yet another embodiment of the invention is illustrated in FIG. 3. Using an SRAM as the interface between memory core/sense amplifiers and data path circuitry further enhances random access latency and bus turn around time. FIG. 3 illustrates such a preferred embodiment bank architecture. These SRAM registers can be staging areas for the INGRESS and EGRESS of data (packets or other forms). The Global Address Supervisor interacts with the SRAM registers when the BANK is available to enhance performance further. All addresses/commands/controls are from SRAM registers, as shown in FIG. 3. When the memory core is flash, such SRAM registers are already available (see references) so that this invention enhances performance without additional cost to the user.

[0033] Yet another embodiment is shown in FIG. 4 where the READ and WRITE data paths are separated (individualized). Such separate data path architecture enhances random access performance, as well as double the data throughput (especially for balanced Read/Write schemes in communication applications). The number of SRAM registers (buffers) can be increased, to maintain multiple open pages as well.

[0034] While particular embodiments of the inventions have been shown and described, changes and modifications may be made there in without departing from the inventions in their 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 basic invention.

Claims

1. A memory system comprising at least one integrated circuit with address pipelining to enhance data throughput The addresses toggled on both rising and falling edges of the clock.

2. A memory integrated circuit having at least one bank of memory core, with pipelined addresses supervised by a Global Address Supervisor.

3. The Global Address Supervisor in the integrated circuit having the ability to channel and pipeline addresses to a multibank component having both Active and READY banks.

4. The memory core in the integrated circuit consisting of memory storage cells that are dynamic requiring periodic data refresh.

5. The memory core in the integrated circuit consisting of memory storage cells that are static and not requiring periodic data refresh.

6. The memory core in the integrated circuit consisting of storage cells that a re nonvolatile.

7. The memory core in the integrated circuit consisting of any combination of dynamic, static and nonvolatile memory storage cells.

8. The memory core in the integrated circuit consisting of memory sub-units or arrays or memory cells organized as rows×columns.

9. The memory integrated circuit consisting of page mode operation where a group of columns in a given row are stored in a page.

10. The memory integrated circuit consisting of a prefetched data burst operation in a given page.

11. The memory integrated circuit described in claim 2 capable of operation in a rail-to-rail switching of input and output pins including addresses, data, and clocks

12. The memory integrated circuit described in claim 2 capable of operation in a bus environment

13. The memory integrated circuit described in claim 2 capable of operation in a peer-to-peer environment.

14. The memory integrated circuit described in claim 2 operable in a synchronous clock environment.

15. The memory integrated circuit described in claim 2 operable in an asynchronous clock environment.

16. A SIP (System In Package) consisting of at least one integrated circuit having one embodiment described in claim 2.

17. A SOC (System On a Chip) consisting of at least one embodiment described in claim 2.