High speed memory-based buffer and system and method for use thereof

[0002] 1. Field of the Invention

[0003] This invention relates to electronic devices, and more particularly to a high-speed memory-based buffer, and system and method for use thereof.

[0004] 2. Description of the Related Art

[0005] Semiconductor processing has reached the level of placing tens of millions of transistors on a single integrated circuit or monolithic substrate, also known as a chip. The drive towards miniaturization is evident from recent forecasts from chipmakers towards shrinking transistor size from 0.25 micron down to 0.18 micron and beyond in the next decade. It is anticipated that this trend towards increased miniaturization will continue. This drive towards integration of electronic devices on increasingly complex chips requires increasingly compact designs even for relatively simple electronic devices. A companion concern to size is power consumption, as a high power device compacted into a smaller area may cause heating problems on the chip, as well as the possibility of actually melting the chip during operation.

[0006] Hardware buffers are devices that receive bits, store the bits temporarily, and then provide the bits upon request thereof. Buffers are often shared by a number of other devices that operate at different speeds or with different sets of priorities. The buffer allows faster devices to operate on the data stream without being held up by slower devices. Buffers are often categorized by the order in which the buffer provides data as compared to the order in which the data are stored. Examples of buffers include first in, first out buffers, also known as FIFOs; last in, first out buffers, as known as LIFOs or stacks, as well as queues, reorder buffers, reservation stations; etc. Another important factor is the size of the buffer, including the depth (i.e. the number of different data entries which can be stored concurrently) as well as the width (i.e. the number of bits that can be comprised in a single data entry).

[0007] Hardware buffers are often comprised of memory cells or registers. Memory cells may be comprised of any type of memory, including random access memory (RAM) in any of its varieties. Some memory types are fast on the write-in to read-out turnaround, having high power consumption and a large footprint, i.e. they require a large number of basic components and therefore a large area of semiconductor on a wafer. Other memory types have low power consumption and take up little space but are relatively slow. Registers usually fall into the same category as fast memory. When a buffer is comprised of registers or fast memory, the depth and width of the buffer is severely constrained due to the power consumption and footprint. When a buffer is comprised of slow memory, the speed of the buffer is severely constrained.

[0008] It would be desirable to have a buffer with a small footprint on the chip and low power consumption, yet has a rapid turnaround time for data entries written into it. Preferably, the turnaround time should have no latency, that is, data entries should be accepted into the buffer on each clock cycle until the buffer can no longer accept data entries and data entries should be available to be read out on each clock cycle until the buffer is empty. The buffer should also be relatively deep and wide as a function of the footprint and power consumption.

[0009] The problems outlined above are in large part solved by a system and method for buffering data entries using a memory and one or more registers, where control logic controls the operations of the memory and the registers. The memory may provide for relatively deep and wide storage of data entries, while minimizing footprint and power consumption. The registers may ensure that a data entry may be written in and/or read out on each clock cycle.

[0010] In one embodiment, a buffer configured to store a plurality of data entries comprises a memory, one or more registers coupled between an output of the memory and an output of the buffer, and control logic coupled to the registers and to the memory. The memory and each of the registers are configured to store data entries in data entry storage locations. The control logic is configured to control the operations of the buffer. The combination of memory with the one or more registers coupled between an output of the memory and an output of the buffer may advantageously result in a small footprint and low power consumption. The speed of the registers is added to the high depth and width to size ratio of the memory.

[0011] In another embodiment, the number of registers in the buffer is determined number of clock cycles of read latency of the memory. An alternative description is that the number of registers in the buffer is determined by the number of clock cycles between the clock cycle when a data entry is requested from the memory and the clock cycle when the data entry is available at the output of the memory. The number of registers may be set to optimize the speed gain from the use of registers, while minimizing the size and power consumption to depth and width ratio. The memory preferably provides the majority of the data entry storage locations.

[0012] In various embodiments, the buffer may be configured as a FIFO buffer or as a LIFO buffer. In one embodiment, the buffer is configured to operate as either a FIFO buffer or a LIFO buffer according to a configuration bit which is set in the control logic.

[0013] A method is likewise contemplated, operating within a buffer including a memory and one or more registers chained from the output of the memory. The method comprises reading a first data entry out of the memory and into an output register, the buffer receiving a read request, and providing the first data entry from the output register in response to the buffer receiving the read request. The method may advantageously use the speed of the registers coupled with the depth of the memory, all within a small footprint with relatively low power consumption. The method may further include reading a second data entry out of the memory and into the output register in response to providing the first data entry from the output register. It is noted that this further embodiment primes the output register in preparation for an additional read request during a following clock cycle.

[0014] In one embodiment, the method operates in a buffer including at least two registers, the output register, and an intermediate register coupled between the memory and the output register. The method then may further comprise reading a second data entry out of the memory and into the output register in response to providing the first data entry from the output register. In this embodiment, the method further reads a third data entry out of the memory and into the intermediate register after reading the second data entry out of the memory and into the output register.

[0015] A system is further contemplated, comprising, in one embodiment, a plurality of buffers coupled between one or more output ports and a plurality of input ports. The output ports are coupled to receive data from the input ports. The buffers may be of any embodiment described herein, but are preferably FIFO buffers, with each of the buffers configured to accept data from a respective input port. The control logic of each buffer is configured to output an empty control signal when the buffer is empty of data entries. Input multiplexers are coupled to each of the input ports and to an output of the respective buffer. The empty control signal from the respective buffer selects between data from the respective input port and the respective buffer. An output multiplexer is coupled to receive data from the input multiplexers and to provide output data to a respective output port. The depth of the buffers may advantageously provide for better buffering of data in the system without using unnecessary space or power.

[0016] Other objects and advantages of the invention will become apparent upon reading the following detailed description and upon reference to the accompanying drawings in which:

[0017] FIG. 1 is a block diagram of an embodiment of a buffer using control logic to control the operations of the buffer;

[0018] FIG. 2 is a block diagram of an embodiment of the buffer of FIG. 1 configured as a first in, first out (FIFO) buffer, with both memory and registers as storage locations;

[0019] FIG. 3 is an embodiment of a state machine that controls the operation of the FIFO buffer of FIG. 2;

[0020] FIGS. 4A, 4B, and 4C are timing diagrams illustrating various operations of the FIFO buffer of FIG. 2 as controlled by the state machine of FIG. 3;

[0021] FIG. 5 is a block diagram of an embodiment of the buffer of FIG. 1 configured as a last in, first out (LIFO) buffer, with both memory and registers as storage locations;

[0022] FIG. 6 is a block diagram of an embodiment of the buffer of FIG. 1 configurable as either a FIFO buffer or a LIFO buffer, with both memory and registers as storage locations; and

[0023] FIG. 7 is a block diagram of an embodiment of a system that uses a plurality of the buffers of FIG. 2 to route data between input and output ports.

[0024] While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the drawings and detailed description thereto are not intended to limit the invention to the particular form disclosed, but on the contrary, the intention is to cover all modifications, equivalents and alternatives falling within the spirit and scope of the present invention as defined by the appended claims.

[0025] Turning to FIG. 1, a block diagram of an embodiment of a buffer 100 using control logic 120 to control the operations of the buffer 100 is illustrated. Data entries are written into the buffer 100 on the DATA₁₃IN line 112. Data entries written into the buffer 100 are stored in storage locations 110. Control logic 120 sends and receives control input/output (I/O) signals 122. The control logic 120 also outputs control signals to the storage locations 110. Data entries are read out of the buffer 100 on the DATA₁₃OUT line 114.

[0026] The storage locations 110 may include memory, registers, or other storage elements that operate to accept, maintain, and yield data entries to read requests. Control logic 120 includes one or more combinatorial logic circuits and/or state machines, which operate upon one or more input signals, such as those included in the control I/O 122, to produce one or more output control signals. The output control signals may be used to control various other devices, such as the storage locations 110, as shown. It is noted that the buffer 100 receives and operates according to a clock signal that is not shown.

[0027] FIFO Buffer

[0028] Turning now to FIG. 2, a block diagram of an embodiment of the buffer 100 of FIG. 1 configured as a first in, first out (FIFO) buffer 200, with both memory 202 and registers 204 and 206 as storage locations 110 and a finite state machine 220 included in the control logic 120. Data entries are written into the FIFO buffer 200 on the DATA₁₃IN line 212. Data entries that are written into the buffer 200 are stored in either the memory 202, register REG1204, or register REG0206. Data entries written into the FIFO buffer 200 are input to the memory 202 as well as the register REG1204 and the register REG0206. As shown, the inputs are first put through multiplexers 203 and 205, controlled by the state machine 220. The output of the memory 202 is input to both multiplexers 203 and 205. The output of each register REG1204 and REG0206 is fed back to the multiplexer 203 or 205 that feeds the respective register REG1204 or REG0206. The data entries are read out of the FIFO buffer 200 on the DATA₁₃OUT line 214.

[0029] State machine 220 sends and receives control input/output (I/O) signals 222, including, as shown, a READ signal, a WRITE signal, a FULL signal, and an EMPTY signal. The READ and WRITE signals are used to read and write, respectively, from the FIFO buffer 200. The FULL and EMPTY signals are output to allow an external device to know when the FIFO buffer 200 is full or empty of data entries. The state machine 220 also outputs control signals to the memory 202, including incrementing the read pointer 207 and the write pointer 208. As noted above, the state machine 220 outputs control signals to the multiplexers 203 and 205, including selecting which input to a respective multiplexer 203 or 205 is output by the respective multiplexer 203 or 205. The state machine 220 increments and decrements one or more counters 240 to keep track of how many data entries are stored in the FIFO buffer 200, either in the memory 202 or the registers REG1204 and/or REG0206. A decode unit 230 accepts the output of the counters 240 and provides the state machine 220 with the count of the data entries as well as outputting the FULL and EMPTY signals.

[0030] The general operation of the FIFO buffer 200, as illustrated, may be broadly summarized as follows. A data entry on the DATA₁₃IN line 212 to the FIFO buffer 200 is input to the memory 202, the register REG1204, and the register REG0206. The data entry is stored in register REG0206 if it is empty. If register REG0206 is full, the data entry is stored in register REG1204. If register REG1204 is also full, the data entry is stored in the memory 202 until it is full. During normal operation, the FIFO buffer 200 will only write in a new data entry when the FULL signal is asserted if the FIFO buffer is also reading out an existing data entry stored in the FIFO buffer 200.

[0031] The registers REG0206 and REG1204 allow the FIFO buffer 200 to have low latency between read and write cycles. The FIFO buffer 200 writes an incoming data entry into the registers 204 or 206 before filling the memory 202. Thus, the FIFO buffer 200 may provide data entries from the registers 204 and/or 206 on the immediately following clock cycle. The memory 202 provides additional data entry storage, albeit in a slower storage space.

[0032] The memory 202 fills the registers 204 and 206 to maintain a low latency. As the memory 202 requires multiple clock cycles to read out a data entry stored in the memory 202, the memory starts a memory read cycle in advance of the clock cycle in which the FIFO buffer will need to read out the data entry from the register REG0206. Thus, the memory 202 provides deep storage and the registers 204 and 206 provide ready access to the next data entry.

[0033] Data entries are read out of the FIFO buffer 200 from the register REG0206. The state machine 220 is configured to keep register REG0206 filled from the register REG1204, if possible. Data entries stored in the memory 202 may also be used to keep register REG0206 filled, as will be seen below with respect to the operation of the state machine 220 in FIG. 3 and the timing diagrams of FIGS. 4A, 4B, and 4C.

[0034] In a preferred embodiment, the memory 202 is comprised of synchronous random access memory (RAM), or an asynchronous RAM without an exit flop. In this embodiment, the register REG1204 latches each data output to it by the memory 202. It is noted that the memory 202 may also comprise content addressable memory (CAM) or another memory type as desired. It is also noted that the memory 202 may be addressed, instead of using read pointer 207 and write pointer 208, as desired. In a preferred embodiment, the FIFO buffer 200 is implemented inside an application specific integrated circuit (ASIC) as part of a system that includes the buffering between input and output ports as shown below in FIG. 7.

[0035] State Machine in the Control Logic

[0036] Turning now to FIG. 3, an embodiment of the state machine 220 that controls the operation of the FIFO buffer 200 of FIG. 2 is shown. The states of the state machine 220 include the following. The REG0EMPTY state is indicated when register REG0 is empty, along with the remainder of the FIFO buffer 200. The REG0FULL state is indicated when register REG0 is filled, but register REG1 is empty. The REG1FULL state indicates that both registers REG0 and REG1 are full. The MEMSTART state indicates that at least one data entry is stored in the memory. The BYPASSREG1 state indicates that the register REG1 is empty and that the memory will provide a data entry directly to the register REG0 unless register REG0 is still filled. The MEMGENERAL state indicates the memory includes a plurality of entries, while the system is not in a position to bypass the register REG1.

[0037] With the FIFO buffer 200 empty, the state machine starts in the REG0EMPTY state 310. The state machine loops in the REG0EMPTY state 310 while the FIFO buffer 200 is reading data or not writing data (RD₁₃DATA ∥!WR₁₃DATA is TRUE). When the FIFO buffer 200 is writing data and not reading data (!RD₁₃DATA & WR₁₃DATA is TRUE) in the REG0EMPTY state 310, the state machine transitions to the REG0FULL state 320.

[0038] The state machine loops in the REG0FULL state 320 while the FIFO buffer 200 is reading data while writing data (RD₁₃DATA & WR₁₃DATA is TRUE) or while the FIFO buffer 200 is not reading data and not writing data (!RD₁₃DATA & !WR₁₃DATA is TRUE). If the FIFO buffer 200 reads data but does not write data (RD₁₃DATA & !WR₁₃DATA is TRUE) in the REG0FULL state 320, the state machine transitions to the REG0EMPTY state 310. If the FIFO buffer 200 writes data but does not read data (!RD₁₃DATA & WR₁₃DATA is TRUE) in the REG0FULL state 320, the state machine transitions to the REG1FULL state 330.

[0039] The state machine loops in the REG1FULL state 330 while the FIFO buffer 200 reads data while writing data (RD₁₃DATA & WR₁₃DATA is TRUE) or while the FIFO buffer 200 neither reads data nor writes data (!RD₁₃DATA & !WR₁₃DATA is TRUE). If the FIFO buffer 200 reads data but does not write data (RD₁₃DATA & !WR₁₃DATA is TRUE) in the REG1FULL state 330, the state machine transitions to the REG0FULL state 320. If the FIFO buffer 200 writes data but does not read data (!RD₁₃DATA & WR₁₃DATA is TRUE) in the REG1FULL state 330, the state machine transitions to the MEMSTART state 340.

[0040] The state machine loops in the MEMSTART state 340 while the FIFO buffer 200 neither reads data nor writes data (!RD₁₃DATA & !WR₁₃DATA is TRUE). If the FIFO buffer 200 reads data (RD₁₃DATA is TRUE) in the MEMSTART state 340, the state machine transitions to the BYPASSREG1 state 350. If the FIFO buffer 200 writes data but does not read data (!RD₁₃DATA & WR₁₃DATA is TRUE) in the MEMSTART state 340, the state machine transitions to the MEMGENERAL state 360.

[0041] The state machine loops in the BYPASSREG1 state 350 while the FIFO buffer 200 reads data while writing data and there are not exactly two data entries in the FIFO buffer 200 (RD₁₃DATA & WR₁₃DATA & !TWO₁₃ITEM is TRUE) or while the FIFO buffer 200 reads data and there are not exactly one or two data entries in the FIFO buffer 200 (RD₁₃DATA & !ONE₁₃ITEM & !TWO₁₃ITEM is TRUE). If the FIFO buffer 200 reads data but does not write data and there is not exactly one data entry in the FIFO buffer 200 but there are exactly two data entries (RD₁₃DATA & !WR₁₃DATA & !ONE₁₃ITEM & TWO₁₃ITEM is TRUE), the state machine transitions to the REG0FULL state 320. If the FIFO buffer 200 writes data but does not read data (!RD₁₃DATA & WR₁₃DATA is TRUE) or does not read data and there are not exactly two or three data entries in the FIFO buffer 200 while the state machine is in the BYPASSREG1 state 350, the state machine transitions to the MEMGENERAL state 360. If the FIFO buffer 200 reads data but does not write data and there is exactly one data entry in the FIFO buffer 200 (RD₁₃DATA & !WR₁₃DATA & ONE₁₃ITEM is TRUE) while the state machine is in the BYPASSREG1 state 350, the state machine transitions to the REG0EMPTY state 310.

[0042] In the MEMGENERAL state 360, the state machine loops if the FIFO buffer 200 neither reads nor writes data (!RD₁₃DATA & !WR₁₃DATA is TRUE), or if the FIFO buffer 200 does not read data and is not completely empty of data entries (!RD₁₃DATA & !EMPTY is TRUE), or if the FIFO buffer 200 reads data while it is completely full of data entries and the read pointer and the write pointer point to the same data entry storage in the memory 202 (RD₁₃DATA & FULL & RD₁₃EQ₁₃WR is TRUE). If the FIFO buffer 200 reads data but does not write data and is not completely empty of data entries (RD₁₃DATA & !WR DATA & !EMPTY is TRUE), or if the FIFO buffer 200 reads data while writing data and is not completely full of data entries (RD₁₃DATA & WR₁₃DATA & !FULL is TRUE), or if the FIFO buffer 200 reads data while writing data and the read pointer and the write pointer do not point to the same storage location in the memory 202 (RD₁₃DATA & WR₁₃DATA & !RD₁₃EQ₁₃WR is TRUE), while the state machine is in the MEMGENERAL state 360, the state machine transitions to the BYPASSREG1 state 350. If the FIFO buffer 200 writes but does not read a data entry while empty (!RD₁₃DATA & WR₁₃DATA & EMPTY is TRUE), or if the FIFO buffer 200 reads but does not write a data entry while empty (RD₁₃DATA & !WR₁₃DATA & EMPTY is TRUE), in the MEMGENERAL state 360, the state machine transitions to the REG0EMPTY state 310.

[0043] Other implementations of state machine 220 are contemplated, as well as other forms of control logic 120 configured to operate buffer 200. Some or all of the conditions and/or transitions given for state machine 220 may be deleted, or new states, conditions and/or transitions may be added to achieve operation of the FIFO buffer 200.

[0044] Timing Diagrams

[0045] Turning now to FIGS. 4A, 4B, and 4C, exemplary timing diagrams 400A, 400B, and 400C, respectively, of the FIFO buffer 200 of FIG. 2 operating according to the state machine 220 of FIG. 3 are illustrated. In each case, the clock cycle designations for the diagram 400 are followed by the READ and WRITE signals of control I/O signals 222, the data that is on the DATA₁₃IN 112 and DATA₁₃OUT 114 lines, the contents of the registers REG0206 and REG1204, the contents of the memory 202, the state of the state machine 220, and the positions of the read and write pointers. Memory 20 includes storage locations referred to MEM0, MEM 1, etc. in FIGS. 4A-4C. The depth of the buffer is shown for illustrative purposes only. The memory may include additional data entry storage locations.

[0046] Each clock cycle is a snapshot in time, seen between the rising edges of the clock. Thus, the operations of the buffer transition on the rising edge, with the data moving, the state machine transitioning, and the read and write pointers transitioning on the next rising edge. It is noted that other clock conditions other than the rising edge are well known in the art. It is also noted that a number of registers other than two are contemplated, based on the characteristic latency of the memory used.

[0047] Timing diagram 400A, shown in FIG. 4A, includes eleven clock cycles. The buffer starts out empty, with no data stored in the registers REG0 or REG1 or the memory data entry storage locations MEM0, MEM1, etc. The read and write pointers both point to memory location MEM0. A write cycle is indicated with data entry D0 on the DATA₁₃IN line. The state machine is in the REG0EMPTY state.

[0048] During clock cycle 2, data D0 is now stored in register REG0 and available on the DATA₁₃OUT line. The state machine is in the REG0FULL state. As no data entries are in the memory, the read and write pointers continue to point to MEM0. Both a read and a write cycle are indicated, with data D1 presented on the DATA₁₃IN line.

[0049] During clock cycle 3, data DO has been read out of the buffer, and data D1 has taken its place in register REG0 and on the DATA₁₃OUT line. Read and write cycles are again indicated with data D2 on the DATA₁₃IN line. The state machine stays in the REG0FULL state.

[0050] In clock cycle 4, only a write cycle is indicated. Data D3 is presented on the DATA₁₃IN line. Data D2 is now stored in the register REG0 and available to be latched on the DATA₁₃OUT line. The state machine is currently in the REG0FULL state, and will transition to the REG1 FULL state due to the updating of register REG1 with data D3.

[0051] Both read and write cycles are indicated with data D4 on the DATA₁₃IN line in clock cycle 5. Data D2 is stored in register REG0 and is latched from the DATA₁₃OUT line. Data D3 is stored in the register REG1 and will be latched into register REG0. The state machine has transitioned into the REG1 FULL state and will remain in this state.

[0052] The state machine stays in the REG1FULL state with both a read and a write indicated in clock cycle 6. Data D5 is present on the DATA₁₃IN line to be written in. Data D3 is in register REG0 to be read out on the DATA₁₃OUT line. Data D4 is stored in the register REG1 and will be latched into the register REG0.

[0053] During clock cycle 7, a write cycle occurs with data D6 on the DATA₁₃IN line. Data D4 is stored in register REG0 and available on the DATA₁₃OUT line. Data D5 is stored in the register REG1. The state machine is in the REG1FULL state until the data D6 written in is stored in the memory at data entry storage location MEM0. The write pointer transitions on the write cycle into memory.

[0054] During clock cycle 8, a read cycle occurs with data D4 being read out of register REG0 on the DATA₁₃OUT line. Data D5 is stored in the register REG1 and will be latched into register REG0 on the read cycle. Data D6 is stored in the memory at storage location MEM0. The state machine in the MEMSTART state. The read pointer transitions on the read cycle from memory, as data D6 is being read from memory to replace data D5 in register REG1. The write pointer is now pointing to MEM1.

[0055] As the memory cannot read out data D6 immediately, during clock cycle 9, register REG1 is empty, and the state machine transitions to the BYPASSREG1 state. Data D6 is still in the memory at storage location MEM0. Data D5 is stored in register REG0 and is available on the DATA₁₃OUT line for the indicated read cycle. Both read and write pointers now point to storage location MEM1.

[0056] Another read cycle is indicated during clock cycle 10. Data D6 is now available in the register REG0 and on the DATA₁₃OUT line. The state machine has transitioned to the REG0FULL state.

[0057] During clock cycle 11, the buffer is empty. The state machine is in the REG0EMPTY state, since data D6 has been read out of the buffer.

[0058] Turning now to FIG. 4B, timing diagram 400B includes nine clock cycles. In clock cycle 1, the state machine starts out in the MEMSTART state. Data D0 is stored in register REG0 and available to be read out on the DATA₁₃OUT line. Data D1 is stored in register REG1, and data D2 is stored in the memory in storage location MEM0. The read pointer points to storage location MEM0. The write pointer points to storage location MEM1 and will transition on the indicated write cycle. Data D3 is presented on the DATA₁₃IN line to be written in.

[0059] In clock cycle 2, the state machine has transitioned to the MEMGENERAL state. Data D0 is still stored in register REG0 and available to be read out on the DATA₁₃OUT line. Data D1 is still stored in register REG1. Data D2 is still stored in the memory in storage location MEM0, while data D3 has been stored in the memory in storage location MEM1. The read pointer still points to storage location MEM0, but the write pointer now points to storage location MEM2 and will transition on the indicated write cycle. Data D4 is presented on the DATA IN line to be written in.

[0060] During clock cycle 3, a read cycle is indicated. Data D0, stored in register REG0, will be latched on the DATA₁₃OUT line. Data D1, still stored in register REG1, will be written into register REG0. Data D2, stored in the memory in storage location MEM0, will be read from the memory. As with data D6 in clock cycle 8 of timing diagram 400A, data D2 will remain in the memory until the memory latency period has passed. Data D3 is still stored in the memory in storage location MEM1. Data D4 is now stored in the memory in storage location MEM2. The state machine remains in the MEMGENERAL state. The read pointer points to storage location MEM0, but will transition on the rising edge due to the read cycle. The write pointer now points to MEM3.

[0061] Register REG1 is empty during clock cycle 4. Data D1 is now stored in the register REG0 and available on the DATA₁₃OUT line. Data entries D2, D3, and D4 are all stored in memory at storage locations MEM0, MEM1, and MEM2, respectively. Data entry D2 will be read from memory and bypass into register REG0 during clock cycle 5, since a read is being performed in clock cycle 4. The state machine has transitioned to the BYPASSREG1 state. The read pointer points to the MEM1 state and will transition on the read cycle. The write pointer still points to the MEM3 storage location. A memory read cycle for data D4 is also initiated during clock cycle 4 in anticipation of a read from the buffer during clock cycle 5.

[0062] Another read cycle occurs during clock cycle 5. Data D2 will be read from the register REG0 on the DATA₁₃OUT line. Register REG1 remains empty, and the state machine remains in the BYPASSREG1 state. Data entries D3 and D4 remain stored in memory at storage locations MEM1 and MEM2, respectively, but data entry D3 will be written into the register REG0 during clock cycle 6. The read pointer points to the MEM2 storage location and will transition on the read. The write pointer still points to the MEM3 storage location.

[0063] Another read cycle occurs during clock cycle 6. Data D3 will be read from the register REG0 on the DATA₁₃OUT line. Register REG1 remains empty, and the state machine remains in the BYPASSREG1 state. Data entry D4 remains stored in memory at storage location MEM2, but will be written into the register REG0 during clock cycle 7. Both read and write pointers now point to storage location MEM3.

[0064] During clock cycle 7, data D4 is stored in register REG0 and is available on the DATA₁₃OUT line. A write cycle is indicated. Data D5 is present on the DATA₁₃IN line and will be written directly into register REG1. The state machine has transitioned to the REG0FULL state.

[0065] A write cycle is indicated during clock cycle 8. Data D6 is presented on the DATA₁₃IN line. Data D4 is stored in register REG0 and is available on the DATA₁₃OUT line. Data D5 is stored in register REG1. The state machine has transitioned to the REG1FULL state. Both read and write pointers now point to storage location MEM3, and the write pointer will transition on the write. With both registers REG0 and REG1 full, data D6 will be written into the memory at storage location MEM3.

[0066] During clock cycle 9, no read or write activity occurs. Data D4 remains stored in register REG0 and available on the DATA₁₃OUT line. Data D5 remains stored in register REG1, while data D6 is stored in the memory at storage location MEM3. The state machine has transitioned to the MEMSTART state. The read pointer points to the MEM3 storage location, while the write pointer points to the MEM4 storage location.

[0067] Turning now to FIG. 4C, timing diagram 400C includes eight clock cycles. In clock cycle 1, the state machine starts out in the REG1FULL state. Data D0 is stored in register REG0 and is available to be read out on the DATA₁₃OUT line. Data D1 is stored in register REG1. The read and write pointers both point to storage location MEM0. A read cycle is indicated and the buffer will read out data D0.

[0068] During clock cycle 2, a write cycle is indicated. Data D2 is presented on the DATA₁₃IN line. Data D1 is stored in register REG0 and available to be read out on the DATA₁₃OUT line. The state machine has transitioned to the REG0FULL state.

[0069] During clock cycle 3, another write cycle is indicated. Data D3 is presented on the DATA₁₃IN line. Data D1 is stored in register REG0 and available to be read out on the DATA₁₃OUT line. Data D2 is stored in register REG1. The state machine has transitioned to the REG1FULL state. As data D3 will be written into memory, into storage location MEM0, the write pointer will transition to point to MEM1 on the write.

[0070] In clock cycle 4, both read and write cycles are indicated. Data D4 is presented on the DATA₁₃IN line. Data D1 is stored in register REG0 and will be read out on the DATA₁₃OUT line. Data D2 is stored in register REG1 and will latch into register REG0 as data D1 is read out. Data D3 is stored in the memory at storage location MEM0. A memory read for data D3 is started on the read request. The state machine has transitioned to the MEMSTART state. As both a memory read and a memory write are being performed, the read and write pointers will both transition from their current positions, pointing at storage locations MEM0 and MEM1, respectively.

[0071] In clock cycle 5, again both read and write cycles are indicated. Data D5 is presented on the DATA₁₃IN line. Data D2 is stored in register REG0 and will be read out on the DATA₁₃OUT line. Due to memory latency, no data is stored in register REG1. Data D3 is stored in the memory at storage location MEM0. Data D4 is stored in the memory at storage location MEM1. The memory read for data D3 will complete on the next rising edge, allowing data D3 to be latched into the then empty register REG0. The state machine has transitioned to the BYPASSREG1 state. A memory read for data D4 is started on the read request. As both a memory read and a memory write are being performed, the read and write pointers will both transition from their current positions, pointing at storage locations MEM1 and MEM2, respectively.

[0072] A third consecutive clock cycle with both read and write cycles occurs in clock cycle 6. Data D6 is presented on the DATA₁₃IN line. Data D3 is stored in register REG0 and will be read out on the DATA₁₃OUT line. Due to memory latency, no data is stored in register REG1. Data D4 is stored in the memory at storage location MEM1. Data D5 is stored in the memory at storage location MEM2. The memory read for data D4 will complete on the next rising edge, allowing data D4 to be latched into the then empty register REG0. The state machine remains in the BYPASSREG1 state. A memory read for data D5 is started on the read request. As both a memory read and a memory write are being performed, the read and write pointers will both transition from their current positions, pointing at storage locations MEM2 and MEM3, respectively.

[0073] During clock cycle 7, a write cycle is indicated. Data D7 is presented on the DATA₁₃IN line. Data D4 is stored in register REG0 and available to be read out on the DATA₁₃OUT line. Due to memory latency, no data is stored in register REG1. Data D5 is stored in the memory at storage location MEM2. Data D6 is stored in the memory at storage location MEM3. The state machine remains in the BYPASSREG1 state. The read pointer points at storage location MEM3. The write pointer points to storage location MEM4 and will transition on the write as data D7 is written into memory at to storage location MEM4.

[0074] During clock cycle 8, no I/O cycles are indicated. Data D4 is stored in register REG0 and available to be read out on the DATA₁₃OUT line. Data D5 has been read out of the memory and is stored in register REG1. Data D6 and D7 are stored in memory at storage locations MEM3 and MEM4, respectively. The state machine has transitioned to the MEMGENERAL state. The read pointer points to data D6 in MEM3, while the write pointer points to storage location MEM5.

[0075] LIFO Buffer

[0076] Turning now to FIG. 5, a block diagram of another embodiment of the buffer 100 of FIG. 1 is shown. The buffer 100 is shown configured as a last in, first out (LIFO) buffer 500, with both memory 502 and registers 504 and 506 as storage locations 110 and a finite state machine 520 included in the control logic 120. Data entries are written into the LIFO buffer 500 on the DATA₁₃IN line 512. Data entries that are written into the buffer 500 are stored in either the memory 502, register REG1504, or register REG0506. Data entries that are written into the FIFO buffer 500 are input to the register REG0506. As shown, the inputs are first put through multiplexers 503 and 505, controlled by the state machine 520. The output of the memory 502 is input to both multiplexers 503 and 505. The output of each register REG1504 and REG0506 is fed back to the multiplexer 503 or 505 that feeds the respective register REG1504 or REG0506. The data entries are read out of the LIFO buffer 500 on the DATA₁₃OUT line 514.

[0077] State machine 520 sends and receives control input/output (I/O) signals 522, including, as shown, a READ signal, a WRITE signal, a FULL signal, and an EMPTY signal. The READ and WRITE signals are used to read and write, respectively, from the LIFO buffer 500. The FULL and EMPTY signals are output to allow an external device to know when the LIFO buffer 500 is full or empty of data entries. The state machine 520 also outputs control signals to the memory 502, including incrementing the read pointer 507 and the write pointer 508. As noted above, the state machine 520 outputs control signals to the multiplexers 503 and 505, including selecting which input to a respective multiplexer 503 or 505 is output by the respective multiplexer 503 or 505. The state machine 520 increments and decrements one or more counters 540 to keep track of how many data entries are stored in the LIFO buffer 500, either in the memory 502 or the registers REG1504 and REG0506. A decode unit 530 accepts the output of the counters 540 and provides the state machine 520 with the count of the data entries as well as outputting the FULL and EMPTY signals.

[0078] The general operation of the LIFO buffer 500, as illustrated, may be broadly summarized as follows. A data entry on the DATA₁₃IN line 512 to the LIFO buffer 500 is input to the register REG0506. The data entry is stored in register REG0506. If register REG0506 was already full, the data entry previously stored in register REG0506 is stored in register REG1504. If register REG1504 was also already full, the data entry stored in register REG1504 is stored in the memory 502 until the memory 502 is full.

[0079] Data entries are read out of the LIFO buffer 500 from the register REG0506. The state machine 520 is configured to keep register REG0506 filled from the register REG1504, if possible. Data entries stored in the memory 502 may also be used to keep register REG0506 filled, in a similar fashion as seen above with respect to the operation of the state machine 220 in FIG. 3 and the timing diagrams of FIGS. 4A, 4B, and 4C, both with respect to the FIFO buffer 200.

[0080] It is noted that state machine 520 may differ from state machine 220 due to the differences in FIFO and LIFO operation. However, these differences may be accounted for according to the remainder of the disclosure. For example, the READ and WRITE pointers always move together in the LIFO buffer, once a data entry is stored in the memory. The READ pointer always points to the last entry written into memory, while the WRITE pointer always points to the immediately following empty entry, if one is available. Thus the condition where the READ and WRITE pointers point to the same data entry storage location (RD₁₃EQ₁₃WR) is only TRUE when the memory in the LIFO is either empty or full.

[0081] Dual-Use FIFO or LIFO Buffer

[0082] Turning now to FIG. 6, a block diagram of an additional embodiment of the buffer 100 of FIG. 1 is shown which is dynamically configurable as a FIFO buffer 200 or as a LIFO buffer 500. Operation by the configurable buffer 600 as a FIFO buffer 200 or as a LIFO buffer 500 is determined by a configuration bit 621 in the control logic. The configuration bit may be set by strobing the READ and WRITE input control signals or by another write cycle mechanism, as desired.

[0083] Differences between the configurable buffer 600 and the FIFO buffer 200 and the LIFO buffer 500 will be highlighted here. State machine 620 sends and receives control input/output (I/O) signals 622, including, as shown, a READ signal, a WRITE signal, a FULL signal, and an EMPTY signal. The READ and WRITE signals are used to read and write, respectively, from the configurable buffer 600. The state machine 620 also outputs control signals to the memory 602, including incrementing the read pointer 607, the write pointer 608, and asserting the input selection control signal to multiplexer 609. The input selection control signal determines whether the memory 602 inputs the DATA₁₃IN signal 612 or the register REG1602 output signal from the multiplexer 609. As noted above, the state machine 620 also outputs control signals to the multiplexers 603 and 605. The state machine 640 also controls the counter 640 and the operation of the decode unit 630 as described above.

[0084] System for Buffering between Input and Output Ports

[0085] Turning now to FIG. 7, a block diagram of a system including a plurality of buffers 100 to route data between input and output ports is illustrated. The buffers 720A, 720B are preferably similar to FIFO buffer 200, although LIFO buffer 500 or configurable buffer 600 may also be substituted. As shown, dual input/output (I/O) ports 710A and 710B input and output data. I/O ports 710A and/or 710B may also comprise input only ports or output only ports, as desired, with additional input and/or output ports included in the system. The use of dual input/output (I/O) ports is for simplicity only. In a preferred embodiment, ten dual I/O ports 710 are contemplated; each dual I/O port 710 being configured to route data similarly to that shown in FIG. 7.

[0086] Data input at an I/O port 710 are routed to FIFO buffer 720 and to a data multiplexer 730. The EMPTY signal from the FIFO buffer 720 controls the output of the data multiplexer 730. If the FIFO buffer 720 asserts the EMPTY signal, then the data multiplexer 730 outputs the data entry input at the I/O port 710, and the data entry at the I/O port 710 is not stored in the FIFO buffer 720. If the FIFO buffer 720 does not assert the EMPTY signal, then the data multiplexer 730 outputs the data entry output from the FIFO buffer 720, and the data entry at the I/O port 710 is stored in the FIFO buffer 720.

[0087] As shown, all data input at the I/O ports 710, either directly or buffered through the FIFO buffers 720 are routed to the output multiplexers 740. A source selection signal SOURCE₁₃SEL# is asserted to determine which input to the output multiplexer 740 is output to the I/O port 710.

[0088] It is noted that in the above-described embodiments, specific types of buffers 100 are illustrated. The specific choice of operation of the buffer 100, as FIFO buffer 200, LIFO buffer 500, or as configurable buffer 600 may vary in various embodiments. In certain preferred embodiments, buffer 100 is FIFO buffer 200. Similarly, the number of registers may vary form embodiment to embodiment. The number of registers is preferably optimized based on the latency of the memory. In a preferred embodiment, the buffer is configured to read or write a data entry for each clock cycle, without an intervening idle clock cycle. The number of registers required for this embodiment should allow the memory to fill the output buffer every clock cycle, as necessary.

[0089] Numerous variations and modifications will become apparent to those skilled in the art once the above disclosure is fully appreciated. It is intended that the following claims be interpreted to embrace all such variations and modifications.