Classifications

• • G—PHYSICS
  • G11—INFORMATION STORAGE
  • G11C—STATIC STORES
  • G11C11/00—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor
  • G11C11/21—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using electric elements
  • G11C11/34—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using electric elements using semiconductor devices
  • G11C11/40—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using electric elements using semiconductor devices using transistors
  • G11C11/41—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using electric elements using semiconductor devices using transistors forming static cells with positive feedback, i.e. cells not needing refreshing or charge regeneration, e.g. bistable multivibrator or Schmitt trigger
  • G11C11/413—Auxiliary circuits, e.g. for addressing, decoding, driving, writing, sensing, timing or power reduction

Definitions

• the present invention relates to asynchronous digital circuit design and in particular to an asynchronous static random access memory.
• design complexity e.g., very large scale integration (VLSI) devices with 10 million or more transistors
• signal propagation delay has become a dominant design consideration. It has become clear that a significant design paradigm shift will be necessary if digital circuit design is to continue its historical adherence to Moore's law.
• Asynchronous VLSI is an active area of research and development in digital circuit design. It refers to all forms of digital circuit design in which there is no global clock synchronization signal.
• Delay-insensitive asynchronous designs by their very nature are insensitive to the signal propagation delays which have become the single greatest obstacle to the advancement of traditional design paradigms. That is, delay- insensitive circuit design maintains the property that any transition in the digital circuit could have an unbounded delay and the circuit will still behave correctly.
• the circuits enforce sequencing but not absolute timing. This design style avoids design and verification difficulties that arise from timing assumptions, glitches, or race conditions. Generally speaking, synchronous design styles are facing serious performance limitations. Certain asynchronous design methodologies also have difficulties with some of the same types of limitations, e.g., race conditions.
• delay-insensitive branch of asynchronous digital design because of its relative immunity to these limitations, appears to hold great promise for supporting future advancements in the performance of digital circuits.
• delay-insensitive asynchronous digital design please refer to the following papers: AJ. Martin, "Compiling Communicating Processes into Delay-Insensitive Circuits," Distributed Computing, Nol.l, No. 4, pp. 226-234, 1986; UN. Cvunmings, A.M. Lines, AJ. Martin, "An Asynchronous Pipelined Lattice Structure Filter.” Advanced Research in Asynchronous Circuits and Systems, IEEE Computer Society Press, 1994; AJ. Martin, A.M.
• a static random access memory including a plurality of SRAM state elements and SRAM environment circuitry.
• the SRAM environment circuitry is operable to interface with external asynchronous circuitry and to enable reading of and writing to the SRAM state elements in a delay-insensitive manner provided that at least one timing assumption relating to bit lines included in the SRAM environment circuitry is met.
• the at least one timing assumption comprises at least one of assuming sufficient pre-charging of the bit lines in response to an enable signal representing completion of a previous memory access operation, and assuming latching of the SRAM state elements in response to write signals on the bit lines.
• the at least one timing assumption comprises a single timing assumption which assumes latching of the SRAM state elements in response to write signals on the bit lines.
• the SRAM state elements comprise either of conventional six-transistor (6T) SRAM state elements, or conventional ten-transistor (10T) SRAM state elements.
• the SRAM environment circuitry comprises read circuitry, write circuitry and address generation circuitry.
• the address generation circuitry is operable to generate delay- insensitive addresses from an asynchronous address channel.
• the read circuitry is operable to facilitate transmission of a data token from the SRAM state elements to the external asynchronous circuitry via a read channel in response to a read instruction and an enable signal representing completion of a previous operation.
• the write circuitry is operable to facilitate transmission of a data token from the external asynchronous circuitry to the SRAM state elements via a write channel in response to a write instruction and an enable signal representing completion of a previous operation.
• the SRAM environment circuitry is operable to enable reading of and writing to the SRAM state elements in the delay-insensitive manner provided that an additional timing assumption is met, the additional timing assumption being that the address generation circuitry has decoded an address value.
• the SRAM state elements and the SRAM environment circuitry are organized into a plurality of SRAM banks and the external asynchronous circuitry includes a write channel and a read channel.
• the SRAM further includes split circuitry for enabling transmission of write data tokens from the write channel to any of the SRAM banks, and merge circuitry for enabling transmission of read data tokens from any of the SRAM banks to the read channel.
• the external asynchronous circuitry includes a plurality of write channels
• the split circuitry is a crossbar circuit which is operable to route the write data tokens from any of the plurality of write channels to any of the SRAM banks according to routing control information.
• the external asynchronous circuitry includes a plurality of read channels
• the merge circuitry is a crossbar circuit which is operable to route the read data tokens from any of the SRAM banks to any of the plurality of read channels according to routing control information.
• Fig. 1 is a schematic diagram of a single-ported 6T SRAM state element for use with various embodiments of the present invention.
• Fig. 2 is a block diagram of an asynchronous SRAM designed according to a specific embodiment of the invention.
• Figs. 3-5 include transistor and higher level diagrams for read and write circuitry for an SRAM employing a 6T state element according to a specific embodiment of the invention.
• Fig. 6 is a block diagram of an asynchronous SRAM designed according to a specific embodiment of the invention.
• Fig. 7 includes various transistor and higher level diagrams of portions of the SRAM design of Fig. 6.
• Fig. 1 is a schematic diagram of a single-ported 6T SRAM state element for use with various embodiments of the present invention.
• Fig. 2 is a block diagram of an asynchronous SRAM designed according to a specific embodiment of the invention.
• Figs. 3-5 include transistor and higher level diagrams for read and write circuitry for an SRAM employing a 6T state
• FIG. 8 is a block diagram of a particular implementation of a banked SRAM designed according to the present invention.
• Figs. 9-14 include transistor and higher level diagrams of Split and Merge circuits for use with various banked SRAMs designed according to the present invention.
• Fig. 15 is a block diagram of a multi-ported banked SRAM designed according to the present invention which includes circuits to convert to synchronous domains.
• Fig. 16 is a schematic diagram of a dual-ported 10T SRAM state element for use with various embodiments of the present invention.
• Figs. 17-21 include transistor and higher level diagrams for an SRAM employing a 10T state element according to a specific embodiment of the invention.
• the asynchronous design style employed in conjunction with the invention is characterized by the latching of data in channels instead of registers. Such channels implement a FIFO (first-in-first- out) transfer of data from a sending circuit to a receiving circuit.
• FIFO first-in-first- out
• a four-phase handshake between neighboring circuits implements a channel.
• the four phases are in order: 1) Sender waits for high enable, then sets data valid; 2) Receiver waits for valid data, then lowers enable; 3) Sender waits for low enable, then sets data neutral; and 4) Receiver waits for neutral data, then raises enable.
• this handshake protocol is for illustrative purposes and that therefore the scope of the invention should not be so limited.
• data are encoded using lofN encoding or so-called "one hot encoding.”
• This is a well known convention of selecting one of N+l states with N wires.
• the channel is in its neutral state when all the wires are inactive.
• the chamiel is in its kth state. It is an error condition for more than one wire to be active at any given time.
• the encoding of data is dual rail, also called lofl. In this encoding, 2 wires (rails) are used to represent 2 valid states and a neutral state.
• larger integers are encoded by more wires, as in a lo ⁇ or lofA code.
• multiple 1 ofl s may be used together with different numerical significance.
• 32 bits can be represented by 32 lofl codes or 16 lof4 codes.
• the above-mentioned asynchronous design style may employ the pseudo-code language CSP (concurrent sequential processes) to describe high- level algorithms and circuit behavior.
• CSP is typically used in parallel programming software projects and in delay-insensitive VLSI.
• CSP is sometimes known as CHP (for Communicating Hardware Processes).
• CHP Communicating Hardware Processes
• an SRAM is provided which is operable in an asynchronous environment, and which is robust to delay and transistor variations.
• Fig. 1 illustrates a conventional six-transistor (6T) SRAM state element 100 for use with various embodiments of the invention.
• SRAM state element 100 includes a pair of cross-coupled inverters 102 and 104 with transistor pass gates 106 and 108 connected to corresponding bit lines b.O and b.l.
• An address line 110 is operable to turn on pass gates 106 and 108 thereby exposing the bit lines to the states on the internal nodes x.O and x.l of the SRAM state element.
• Bit lines b.O and b.l can be thought of as a dual-rail representation of the data stored in SRAM state element 100. As will be shown, this characteristic of the conventional SRAM state element makes it amenable for interacting with asynchronous channels designed according to a variety asynchronous design styles. According to a specific embodiment, the interaction between a conventional SRAM state element and the asynchronous handshake protocol associated with a specific design style is particularly efficient.
• address line 110 is used to turn on transistor pass gates 106 and 108 to connect bits lines b.O and b.l (which are pre- charged high) to internal nodes x.O and x.l, respectively.
• Fig. 2 is a high level block diagram in which the SRAM transistor and circuit topologies of the present invention may be employed.
• Read/write circuitry 202 associated with an array 204 of SRAM state elements has a write channel 206 going in and a read channel 208 coming out, and is coupled to array 204 via a plurality of bi-directional bit lines.
• each of the read and write channels is dual rail by N bits, and the bit lines are also dual rail (i.e., b.O and b.1) by N bits.
• a demultiplexer 210 receives an instruction I comprising write and read control signals "iw" and "ir” and selects the appropriate address line(s) associated with SRAM array 204.
• the nature of read/write circuitry 202 will now be described with reference to Figs. 3-5.
• circuitry 202 is organized into three levels of hierarchy. The following description will begin on the first and lowest level of hierarchy as represented by read and write circuitry 300 (Fig. 3) comprising write circuits 302 and 304 which are configured to drive one of a pair of bit lines (b.O and b.l) low during a write operation.
• Enable signal "en” pre-charges the bit lines b.O and b.l, signals when the system is ready to write, and is reset every cycle.
• Signal w.O is high when a logic "0" is to be written to the selected SRAM cell(s)
• signal w.l is high when a logic "1" is to be written to the selected SRAM cell(s).
• Read circuitry 306 comprises read circuits 308 and 310. When read control signal "ir" goes high (and therefore _ir goes low), the corresponding transistors in circuits 308 and 310 are turned on.
• bit lines b.O and b.l exposes bit lines b.O and b.l to the internal states x.O and x.l of a corresponding state element.
• the enable signal "en” and the read control signals "ir” are mutually exclusive, thereby preventing a read operation to be enabled while the bit lines are pre-charging.
• circuits 308 and 310 could be replaced with sense amps.
• Fig. 4 shows the second or next level of the hierarchy 400 of read/write circuitry 202 in which read and write circuitry 300 of Fig. 3 is reproduced four times, once for each of four SRAM state elements each of which corresponds to a set of state elements in the SRAM array.
• the state elements in the arrays are divided into sets to reduce the capacitive loading on the state element circuitry. According to various embodiments and depending upon the size of the SRAM array, the number of sets may vary. The embodiment shown is the four sets by one bit version of the circuit.
• the four _ir signals for the four instances of circuitry 306 are mutually exclusive low, i.e., only one can be low at any given time.
• circuits 402 and 404 reset read lines r.0 and r.1 when all of the inverted read selects go high.
• a half buffer circuit 406 may be optionally employed to boost read signals r.0 and r.l to read signals R.0 and R.l, respectively, which buffers the load on r.0 and r.l, and provides a read completion circuit efficiently, i.e., read completion signal R.v.
• Write completion signal _W.v is generated by logically combining signals w.O and w.l with NOR gate 408.
• the generation of _W.v may optionally include a third signal w.2 which indicates not to write that bit and can implement individual bit enables which can be used to mask specific bits.
• the robustness of these circuits in the face of delay variations is predicated on some simple timing assumptions which relate to the driving of the bit lines. That is, when the enable signal "en” is driven low causing the bit lines to precharge, there is no acknowledgement, the assumption being that the bit lines will be precharged in a sufficiently short period of time to ensure otherwise delay-insensitive operation. Similarly, when the bit lines are driven low or high for write operations, there is no acknowledgement, the assumption being that the bit lines and state bits will stabilize in a sufficiently short period of time. These assumptions have considerable margin and eliminate the need for more complicated completion circuitry. For example, these assumptions allow the write completion to be a simple combination of the signals w.O and w.l.
• the demultiplexing of the address channel may also be unacknowledged, an additional timing assumption being made that the demultiplexing operation will occur within a sufficient period of time to allow otherwise delay- insensitive operation.
• Fig. 5 shows the third or next higher level of the hierarchy of read/write circuitry 202 in which the circuitry of Fig. 4 is reproduced four times (represented by blocks 400) to create a four sets by four bits block. And as described above, each of blocks 400 includes four of blocks 300 which are, in turn, each reproduced for each of N bits (e.g., 32 or 64 bits) in a row.
• the instruction "I" received by the demultiplexer e.g., demultiplexer 210 of Fig. 2 identifies which sets are to be selected.
• a lof8 code (LO through 1.7) is used to identify the type of operation (i.e., read or write) and which of the four sets is selected.
• the read selects (LO through 1.3) are inverted (block 504) to get the mutually exclusive low set of signals _ir.O tlirough _ir.3 discussed above.
• the write selects (1.4 through 1.7) are buffered (block 506) to get signals iw.O through iw.3. This buffering approach is suitable for cases where there is sufficient time to accomplish this.
• blocks 504 and 506 may comprise pipelined circuits which are able to latch and acknowledge their respective values more quickly than the simple buffering approach.
• the buffered write selects, write completion signals _W.v[0] tlirough _W.v[3], and read completion signals R.v[0] through R.v[3] are completed (block 512) to generate w.v and r.v.
• the enable signal may be "ramped up" using an inverter chain having an even number of inverters (e.g., block 514) to generate signal EN.
• the signal EN is then combined with each of w.v and r.v using C-elements, the outputs of which are NAND'ed to generate an instruction acknowledge I.e which acknowledges to the instruction channel that the operation has been completed and that another instruction may be issued (block 516).
• write acknowledge signal W.e is generated to acknowledge the write channel (e.g., circuitry 202 of Fig. 2) and to indicate that the write channel is ready for the next token.
• SRAM 600 comprises two substantially identical blocks 602 and 604 on either side of demultiplexer 606 and control circuit 608.
• each of blocks 602 and 604 includes 6 substantially identical rows of the SRAM state elements and the read/write circuitry shown in Figs. 1-5. Different numbers of rows will correspond to different widths, e.g., for a 32-bit wide memory system, each of blocks 602 and 604 would have 4 rows. Each such row (an example of which is shown in the bottom of block 604) includes one instance of the circuitry of Fig.
• SRAM block 600 is a 1.5 kB block which is repeatable to generate a larger SRAM memory system. Details of how demultiplexer 606 maybe implemented to generate address information according to a specific embodiment will now be described with reference to Fig. 7.
• the address information comes in a 3 by lof4 code which is converted to the conventional lofN code for SRAMs (in this case lof64) using 64 3-input AND gates, i.e., NAND gates with an inverter, (block 702).
• the address generation circuitry may be made delay-insensitive using completion circuitry to complete the lofN code.
• completion circuitry may comprise, for example, completion tree 704 (or any equivalent) which generates address completion signal a. v.
• a.v represents completion of the neutrality of the address generation because the read lines themselves complete the validity of the address, i.e., the read could't have occurred without the address.
• a.v represents the validity and neutrality of the address.
• the demultiplexer implementation described above is merely exemplary and should not be used to limit the invention. That is, as long as the lofN code is set and reset appropriately, the demultiplexer may be implemented in a wide variety of ways.
• the instruction acknowledge signals I.e for all of the instances of 4x4 block 610 of Fig. 6 are completed using, for example, eight-input C-element trees or equivalent structures generating a completion signal d.v.
• Inverted versions of the address completion signal a.v, completion signal d.v, and the reset signal are used as input to a circuit 706 which generates the enable signal en which is then broadcast throughout the memory block for all its various uses including as acknowledge for entire control channel, and for all of the lof4 address codes.
• a repair functionality may be introduced into the SRAM blocks.
• a 65 th address line and corresponding state elements are included in the design described l l above.
• the address decoding circuitry may be enabled to perform an address translation so that the 65 th address may be substituted for the detected bad address. This may involve, for example, a stage inserted before the demultiplexer which converts from 64 to 65 addresses. As will be understood, such a capability may also involve alterations to the various completion circuitry to account for this extra circuitry when it is use. According to various other embodiments of the invention, the SRAM design described above may be employed with additional circuitry to implement a banked asynchronous SRAM design. In one such embodiment shown in Fig.
• a plurality of SRAM banks 802 designed as described above employ a Split 804 to select a bank for a write operation via write channel W, and a Merge 806 to select a bank for a read operation via read channel R. These selections are accomplished in response to a control channel IA which includes one read/write bit and n bits of address information.
• Fig. 9 shows a basic block diagram of a Split 900.
• Fig. 10 shows a basic block diagram of a Merge 1000. See also "Pipelined Asynchronous Circuits" by A. Lines incorporated by reference above.
• Split 804 includes one split_env part and Q split_cell parts
• Merge 806 includes one merge_env part and P merge_cell parts.
• each split_cell[i] waits for S to be valid and checks that the value of S equals i (that is, rail S' is true). If so, it checks the enable from its output R[i] e and when that is high, it copies the valid data from L to R[i]. Once the data are copied to R[i], the split_cell[i] lowers its enable to the split_env, se[i].
• a schematic for a split cell 1100 with 1-bit data and 1-bit control (both encoded as lof2 codes) is shown in Fig. 11.
• the split_env tests the validity and neutrality of the L channel, computes the logical AND of the se[0..Q - l]'s from the split_cell's, and produces an acknowledge for the S and L input channels. The validity and neutrality of the S channel is implied by the acknowledges from the split_cell 's.
• a schematic for a split_env 1200 for 1-bit data and 2 split_cell 's is shown in Fig. 12.
• FIG. 13 A schematic for a merge_cell 1300 with 1-bit data and 1-bit control (encoded as lof2 codes) is shown in Fig. 13.
• Each merge_cell[i] waits for M to be valid and checks that the value of M equals i (that is, rail M 1 is true). If so, it waits for a go signal from the merge_env (which includes the readiness of the output enable R e ) and for the input data L[i] to be valid. When this happens, it copies the value of L[i] to R.
• the merge_env checks the validity of R and broadcasts this condition back to all the merge_cells 's by setting rv high. Next, the merge_cell lowers its enables me[i] and L[i] e .
• FIG. 14 A schematic for a merge_env 1400 for 1-bit data and 2 merge cells is shown in Fig. 14.
• the merge_env checks the readiness of the R enable and raises go.
• the M goes directly to the merge_ceH's, one of which responds by setting R to a new valid value.
• the merge_env then raises rv, after which the merge_cell replies with me[i] .
• the merge_env checks the completion of these actions, and then acknowledges M.
• a banked SRAM designed in accordance with the present invention does not employ sense amplifiers, instead allowing the bit lines their full swing. This slows down the operation of a particular bank, but with the appropriate number of SRAM banks and flow control, the effective data rate can be significantly higher than any one bank operating alone.
• Merge 806 is implemented as a full buffer merge circuit which can store one data token on its input and one on its output.
• a crossbar is employed which is operable to route data from any of a plurality of input channels to any of a plurality of output channels according to routing control information.
• Each combination of an input channel and an output channel comprises one of a plurality of links.
• the crossbar is operable to route the data in a deterministic manner on each of the links thereby preserving a partial ordering represented by the routing control information. Events on different links are uncorrelated.
• both Split 804 and Merge 806 may be replaced by a crossbar to increase the ported-ness of the SRAM without making any changes to the basic SRAM state element, h addition the number of ports on either side (i.e., read or write) are not limited to two. However, the practical number will relate to the number of SRAM banks, i.e., if there are only four banks, more than 4 read ports would never be fully utilized. In addition and as shown in Fig.
• SRAM design (and indeed any asynchronous SRAM design within the scope of the invention) may be employed in an otherwise synchronous system using synchronous-to-asynchronous (S2A) conversion circuitry 1506 and asynchronous-to-synchronous (A2S) conversion circuitry 1508 on the write and read ports, respectively.
• S2A synchronous-to-asynchronous
• A2S asynchronous-to-synchronous
• suitable conversion circuitry is described in copending U.S. Patent Application No. 10/212,574 entitled TECHNIQUES FOR FACILITATING CONVERSION BETWEEN ASYNCHRONOUS AND SYNCHRONOUS
• the system may count on the overall SRAM performance equivalent to the bank speed, or in a different possible embodiment, logic could be added to signal a "wait state" to the synchronous system, based on detecting a bank conflict in the asynchronous SRAM, so that the synchronous system can build a condition into its state-machine that encodes "no SRAM access this cycle” and may thus behave “deterministically” and still attain the performance of the asynchronous SRAM ports on the average.
• SRAM blocks are provided which employ conventional ten transistor (10T) dual-ported SRAM state elements.
• 10T state element 1600 is shown in Fig. 16 in which two buses are employed, one for writing and one for reading. That is, write bus lines _w.O and _w.l are used in conjunction with address line iw for writing to state element 1600, while read bus lines _r.O and _r.l are used in conjunction with address line ir for reading from state element 1600.
• write bus lines _w.O and _w.l are used in conjunction with address line iw for writing to state element 1600
• read bus lines _r.O and _r.l are used in conjunction with address line ir for reading from state element 1600.
• the 10T state element requires more die area, but is considerably faster.
• substantially simultaneous read and write operations to different addresses in the same SRAM bank may occur with 10T state elements.
• a dual-ported, 10T embodiment might be useful, for example, in applications where the need for speed overwhelms area concerns, or where the size of the memory (in terms of how much information is intended to be stored) is relatively small (e.g., a register file), hi the latter type of embodiment, it will be understood that, where the memory is small enough, the need to divide the SRAM into sets may be greatly diminished or eliminated.
• both embodiments with sets and embodiments without sets are contemplated.
• FIG. 17-21 A specific implementation of an asynchronous SRAM employing 10T state elements (e.g., state element 1600 of Fig. 16) will now be described with reference to Figs. 17-21. It will be understood by those of skill in the art that because the hierarchical framework and many of the implementation details discussed above with reference to those embodiments is sufficiently similar to the dual-ported embodiments, such details are not discussed again here for the sake of brevity. That is, only the significant differences between the two types of embodiments are discussed. It should also be noted that, according to a specific implementation, 10T embodiments may be more robust to delay variations than previously described embodiments in that such embodiments may be implemented using only a single timing assumption, i.e., that the 10T state elements have latched in response to the write signals.
• 10T embodiments may be more robust to delay variations than previously described embodiments in that such embodiments may be implemented using only a single timing assumption, i.e., that the 10T state elements have latched in response to the write signals.
• the enable signal "en” is employed to pre-charge the read bus (i.e., _r.O and _r.l).
• Read port cell 1702 which comprises a one-transition output half-buffer boosts read signals _r.O and _r.l to read signals R.0 and R.l, respectively, while also providing a read completion signal _r.v.
• a read operation on the conventional 10T SRAM state element is able to produce an output which is in the proper event-driven dual-rail channel format of the asynchronous system.
• read port cell 1702 has two fewer transitions than its counterpart in the single-ported design.
• Write port cell 1704 is also relatively straightforward, simply inverting signals w.O and w.1 and driving the write buses with the output, while also generating write completion signal _w.v.
• completion of all the corresponding signals _w.v (for write operations), or all the corresponding signals _r.v (for read operations) may be accomplished in the same manner as for the 6T embodiments using Mueller C-element trees having log 2 (N) stages where N is the number of information bits.
• address completion may be accomplished in a similar manner, i.e., with an OR tree, as described above.
• a dual-ported 10T embodiment based on the 10T SRAM state element shown in Fig. 16, and including the read and write port cells of Fig.
• Write Ctrl 1802 combines the 4 per bit write completion signal instances of _w.v from write port cells 1704. Write Ctrl 1802 handshakes with the address control through the signal kw, and produces jwe. Write Ctrl 1802 also acknowledges the 4 bit write data W to the larger SRAM, utilizing the write completion timing assumption.
• a specific implementation of a Write Ctrl 1802 is shown in Fig. 19.
• C-element 1804 collects the N write nibble completion signals jwe, and produces the write demux precharge signal jw.e.
• Read Ctrl 1806 combines the 4 per bit read completion signal instances of _r.v from read port cells 1702. Read Ctrl 1806 handshakes with the address control through the signal kr, and produces jre. Read Ctrl 1806 also produces the precharge signal en that precharges the read lines of the SRAM. A specific implementation of Read Ctrl 1806 is shown in Fig. 19.
• C-element 1808 collects the N read nibble completion signals jre, and produces the write demux precharge signal jr.e.
• Demultiplexers 1810 de-multiplex the address and read/write instruction encoded in I A into individual read and write select lines for the 32 lines.
• the read decode may be precharged with jr.e and the write decode may be precharged with jw.e.
• a specific implementation of a Demultiplexer 1810 is shown in Fig. 20.
• gate 1812 completes the 1 of 32 read select creating the signal kr
• Or gate 1813 completes the 1 of 32 write select, creating the signal l w.
• Ctrl 1814 completes IA, and combines IA completion with kr and kw, producing IA.e, the LA channel acknowledge.
• a specific implementation of Ctrl 1814 is shown in Fig. 21.
• SRAM block 1800 includes an array of 10T SRAM state elements 1600 combined with asynchronous four-phase channels.
• conditional and non-conditional asynchronous channels are defined.
• a non-conditional channel goes through a handshake on every cycle, while a conditional channel only cycles under certain control circumstances.
• the instmction channel is non-conditional. It contains an address and a read/write (or read and write) instruction for every access to the memory.
• the write channel is conditional on the write instruction and places data from the W channel into the array.
• the read channel is conditional on the read instruction and gets data out of the array and places it on the R channel.
• the signals kr and kw are the mechanisms used to synchronize the W and R channels with Ctrl 1814.
• An instruction enters the SRAM via channel IA and is decoded into a lof32 read select (ir) line or a lof32 write select line (iw) in Demultiplexer 1810. According to a specific embodiment, for efficiency this decoding is done in a precharged computation.
• the Demux Complete 1812 "completes" the lof32 code (ir for a read, iw for a write).
• the data is buffered in write port cells 1704 and broadcast across the state bits (_w.x).
• Write Ctrl 1802 completes _w.v and combines the completion signal with kw generating it's completion jwe.
• Write Complete 1804 completes the N jwe signals and produces jw.e which, according to one embodiment, is the precharge for Demultiplexer 1810 for the write select. It should be noted that this precharging may be included for performance but is not necessary for a functioning implementation.
• the read select lines allow the state bit to pull down the read lines.
• Read port cells 1702 contain a "pipelined" buffer stage, buffering _r to R. Read port cells 1702 also contain the read line precharge transistors.
• Read Ctrl 1806 completes the read lines, combines that completion with kr and generates the read line precharge signal en, and its own completion signal jre.
• Read Complete 1808 completes the N jre signals and produces jr.e which, according to one embodiment, is used to precharge the decode for the read select line (ir) in Demultiplexer 1810. Again, it should be noted that this precharging may be included for performance but is not necessary for a functioning implementation.
• Ctrl 1814 completes the address and the kr and kw signals to generate the acknowledge for the instruction, IA.e.
• Ctrl 1814 may be implemented to allow simultaneous writes and reads to the SRAM array. That is, as long as it can be guaranteed that the addresses for the read and write instructions will not collide, simultaneous write and read instructions may be executed. Additional circuitry can be added outside the bank of SRAM to detect when the same address is being read and written, and instead of reading from the SRAM, bypass the write data directly to the read port. According to various embodiments, additional read and write ports may be added by duplicating the appropriate portions of the circuitry described above and by including additional read or write buses. According to various embodiments of the invention, each of the bit lines has an associated staticizer to keep the buses at their pre-charged levels when not in use.
• analog circuit techniques are applied to these staticizers which enable the bit lines to be pulled down more quickly.
• the circuits and processes described herein maybe represented (without limitation) in software (object code or machine code), in varying stages of compilation, as one or more netlists, in a simulation language, in a hardware description language, by a set of semiconductor processing masks, and as partially or completely realized semiconductor devices.
• the various alternatives for each of the foregoing as understood by those of skill in the art are also within the scope of the invention.

Landscapes

• Engineering & Computer Science (AREA)
• Microelectronics & Electronic Packaging (AREA)
• Computer Hardware Design (AREA)
• Static Random-Access Memory (AREA)
• Dram (AREA)

Priority Applications (4)

Applications Claiming Priority (2)

Publications (3)

Family

ID=34079380

Family Applications (1)

Country Status (6)

Cited By (1)

Families Citing this family (43)

Citations (3)

Family Cites Families (43)

• 2004
  • 2004-07-13 DE DE602004024683T patent/DE602004024683D1/de not_active Expired - Lifetime
  • 2004-07-13 EP EP04778275A patent/EP1647030B1/en not_active Expired - Lifetime
  • 2004-07-13 US US10/890,816 patent/US7050324B2/en not_active Expired - Fee Related
  • 2004-07-13 JP JP2006520315A patent/JP4904154B2/ja not_active Expired - Fee Related
  • 2004-07-13 AT AT04778275T patent/ATE452408T1/de not_active IP Right Cessation
  • 2004-07-13 WO PCT/US2004/022679 patent/WO2005008672A2/en not_active Ceased
• 2005
  • 2005-08-09 US US11/200,722 patent/US7161828B2/en not_active Expired - Fee Related

Patent Citations (3)

Non-Patent Citations (3)

Also Published As

Similar Documents

Legal Events