One or more aspects of the invention relate generally to a reconfiguration port and more particularly, to a reconfiguration port controller for dynamic reconfiguration of an integrated circuit.
Programmable logic devices (“PLDs”) exist as a well-known type of integrated circuit (“IC”) that may be programmed by a user to perform specified logic functions. There are different types of programmable logic devices, such as programmable logic arrays (“PLAs”) and complex programmable logic devices (“CPLDs”). One type of programmable logic device, called a field programmable gate array (“FPGA”), is very popular because of a superior combination of capacity, flexibility, time-to-market, and cost.
An FPGA typically includes an array of configurable logic blocks (“CLBs”) and programmable input/output blocks (“IOBs”). The CLBs and IOBs are interconnected by a programmable interconnect structure. The CLBs, IOBs, and interconnect structure are typically programmed by loading a stream of configuration data (“bitstream”) into internal configuration memory cells that define how the CLBs, IOBs, and interconnect structure are configured. CLBs and IOBs form the programmable part of an FPGA referred to as the “FPGA fabric”, which is subject to program control of the configuration memory cells.
CLBs and IOBs may be interconnected via widely distributed on-chip (“global”) routing resources or regionally specific on-chip (“local”) routing resources of an FPGA, such as one or more traces. Moreover, both global and local resources may be used to distribute signals, such as clock signals. A global routing resource may be programmatically coupled to another global routing resource or a local routing resource, or a local routing resource may be programmatically coupled to another local routing resource using what is known as a programmable interconnect point (“PIP”). Conventionally, PIPs have been programmed or reprogrammed using an externally provided configuration bitstream to program programmable logic. Some other types of circuitry that may be included in an FPGA are transceivers, digital clock managers (“DCMs”), and memory controllers.
DCMs may be programmed for providing any of a variety of clock signals. For example, clock signals of different frequencies or different phase relationships may be provided from a reference clock input to a DCM. Furthermore, DCMs may be programmed for providing such a variety of clock signals. Conventionally, DCMs have been programmed or reprogrammed using an externally provided configuration bitstream to program programmable logic.
In addition to configuration memory cells, groups of system memory cells, sometimes referred to as block random access memories (“BRAMs”), may be included in an FPGA. Like configuration memory cells, such BRAMs conventionally are formed using a standard six transistor (“6T”) static random access memory (“SRAM”) memory cell. However, known forms of either or both static and dynamic random access memory (“DRAM”) memory cells, as well as magnetoresistive random access memory cells (“MRAM”) and flash memory cells, may be included in FPGAs. Conventionally configuration memory cells, as well as system memory cells, were programmed and reprogrammed using an externally provided configuration bitstream.
FPGAs include transceivers, which may be configured for “single-ended” or “differential” signaling. A more recent trend is to provide high-speed transceivers, such as multi-gigabit transceivers (“MGTs”). Transceivers may be programmed to conform to any of a variety of communication standards by programming communication signaling parameters, such as duty cycle, frequency, and preemphasis, among other known communication signaling parameters. Conventionally, transceivers were programmed and reprogrammed using an externally provided configuration bitstream.
Accordingly, it should be appreciated that there are many circuits in a programmable logic device that may be programmed to provide user defined functionality. Furthermore, modern day programmable logic devices may include one or more other devices, such as one or more digital signal processors and microprocessors, among other known integrated circuit devices. For example, microprocessors may be embedded cores (“hard processors”) or programmed into CLBs (“soft processors”). While instructions for such other devices may reside in embedded memory, such as one or more BRAMs, such other devices were subject to there surroundings, namely, configuration of functional blocks programmed or reprogrammed using an externally provided configuration bitstream.
As mentioned above, conventionally an FPGA is programmed by supplying an external bitstream to configure the FPGA. Classically, once an FPGA was configured, it was seldom reconfigured, including without limitation configuration of resources previously not programmed, during operation. This had at least in part to do with having a relatively slow internal access port (“ICAP”) for reconfiguration. Notably, it should be appreciated that an ICAP conventionally may be used to configure or reconfigure an FPGA, as such an ICAP has access to all of the FPGA fabric for purposes of configuration or reconfiguration. However, an ICAP port runs at approximately one-third or less the frequency of which the FPGA may be run. Further impacting the ability to quickly achieve reconfiguration, an ICAP port has a minimum bit reconfiguration “granularity” of one frame. Thus, for example, if only one bit in a 1296 bit frame had to be changed, all 1296 bits were processed to change the one bit.
Accordingly, it would be desirable and useful to provide an integrated circuit having internal dynamic reconfiguration capability that is substantially faster than that afforded by an ICAP.
An aspect of the invention is a controller in a programmable logic device, the controller comprising: a port interface; a read/write interface; a plurality of flip-flops coupling the port interface to the read/write interface. The port interface is configured to receive a plurality of signals, where a portion of the plurality of signals are pipelined through the plurality of flip-flops responsive to a data clock signal of the plurality of signals.
Another aspect of the invention is a method for dynamic configuration of function block logic, comprising: receiving a data clock signal, a data write signal, a data activation signal, a data address signal, and a data input signal; pipelining the data write signal, the data activation signal, the data address signal and the data input signal responsive to the data clock signal to provide pipelined versions of the data write signal, the data activation signal, the data address signal and the data input signal; and writing to storage elements responsive to the pipelined versions of the data write signal, the data activation signal, the data address signal and the data input signal at a rate which is at least approximately a frequency of the data clock signal.
Yet another aspect of the invention is a method for dynamic configuration of function block logic, comprising: receiving a data clock signal, a data activation signal and a data address signal; pipelining the data activation signal and the data address signal responsive to the data clock signal to provide pipelined versions of the data activation signal and the data address signal; and reading from storage elements responsive to the pipelined versions of the data activation signal and the data address signal at a rate which is at least approximately a frequency of the data clock signal.
Accompanying drawing(s) show exemplary embodiment(s) in accordance with one or more aspects of the invention; however, the accompanying drawing(s) should not be taken to limit the invention to the embodiment(s) shown, but are for explanation and understanding only.
In the following description, numerous specific details are set forth to provide a more thorough description of the specific embodiments of the invention. It should be apparent, however, to one skilled in the art, that the invention may be practiced without all the specific details given below. In other instances, well known features have not been described in detail so as not to obscure the invention.
FPGA 10 includes configurable logic blocks (“CLBs”) 26, programmable input/output blocks (“IOBs”) 22, memory, such as block random access memory 28, delay lock loops (DLLs) and multiply/divide/de-skew clock circuits which collectively provide digital clock managers (DCMs) 13, and multi-gigabit transceivers (“MGTs”) 24. An external memory may be coupled to FPGA 10 to store and provide a configuration bitstream to configure FPGA 10, namely, to program one or more configuration memory cells to configure CLBs 26 and IOBs 22. Notably, IOBs 22, as well as MGTs 24, are disposed in a ring or ring-like architecture forming a perimeter of I/Os around CLBs 26 of FPGA 10.
Additionally, FPGA 10 may include an Internal Configuration Access Port (“ICAP”) 16, an embedded processor 30, and an embedded system monitor 20.
Though FPGA 10 is illustratively shown with a single embedded processor 30, FPGA 10 may include more than one processor 30. Additionally, known support circuitry, for interfacing with embedded processor 30 may be included in FPGA 10. Furthermore, rather than an embedded processor 30, processor 30 may be programmed into configurable logic such as a “soft” processor 30.
Although
FPGA 10 in one embodiment may be configured and reconfigured (after initially being configured) in response to a configuration information (commands and data) bitstream, that is loaded into a configuration memory array of FPGA 10, either externally, from an external memory, e.g., a read-only memory (“ROM”), via configuration interface 14 and configuration logic 12 or internally, via the ICAP 16 which is also connected to the configuration logic 12 (not shown). Configuration interface 14 can be, for example, a select map interface, a Joint Test Action Group (“JTAG”) interface, or a master serial interface.
The ICAP 16 is used for internal or self-reconfiguration of the FPGA 10. For example, after initial configuration of the FPGA by a configuration bit stream being sent by an external PROM (programmable ROM) to the configuration interface 14, the configured FPGA 10 is put in operational use. Next, part of configured FPGA 10 may be reconfigured under control of the embedded processor 30 using the ICAP 16. This self-reconfiguration is further discussed in a co-pending patent application, Ser. No. 10/377,857, entitled “Reconfiguration of a Programmable Logic Device Using Internal Control”, by Brandon J. Blodget, et. al, filed Feb. 28, 2003, which is incorporated by reference herein in its entirety.
FPGA 10 in another embodiment of the present invention can receive the configuration bitstream via the configuration interface 14 (external configuration or reconfiguration), ICAP 16 (internal reconfiguration), or in addition, by one or more dynamic reconfiguration ports (not shown). Each dynamic reconfiguration port (DRPORT) directly addresses its own group of configuration memory cells (via a controller) for internal reconfiguration without going through the configuration logic 12. This is different than using configuration interface 14 or ICAP 16, which must go though the configuration logic 12 to get to the configuration memory cells.
With renewed reference to
CLB tiles 43 are laid out in a two-dimensional array. In this example, each CLB tile 43 includes a portion of a programmable interconnect structure such that at least part of the programmable interconnect structure for FPGA 50 is formed by the various portions of the many CLBs when CLB tiles 43 are formed together for FPGA 50. Also illustrated are block random memory/multiplier (BRAM/Multiplier) tiles 44.
In order to provide input/output circuitry for interfacing FPGA 50 to external logic, IOB tiles 42 are provided in, e.g., rows 46, 42A and 42B of FPGA 50. In this particular example, an input/output interconnect tile (IOI tile) is used to couple an IOB tile to a CLB tile. Reference numeral 41 points to one such IOI tile. IOI tile 41 is disposed between an IOB tile 42 and a CLB tile 43.
Digital Signal Processors (“DSPs”) are placed in tile area 45. A generally central tile area 46 may be used for support circuitry. The support circuitry may include, for example, DCMs, CCMs, IOBs, configuration logic 55, encryption/decryption logic, global clock driver circuitry, boundary scan circuitry and system monitor 20.
In this particular example, clock distribution circuitry including the backbone of the global clock tree glck 58 is located in area 48. The area 54 represents the bottom half of FPGA 50, which is shown in greater detail in
Additional details regarding FPGA 50 may be found in a co-pending patent application Ser. No. 10/683,944, entitled “Columnar Architecture”, by Steven P. Young, filed Oct. 10, 2003 which is incorporated by reference herein in its entirety.
With continuing reference to
The system monitor may include an analog-to-digital converter (ADC) to monitor parameters like temperature and voltage both on-chip and off-chip. The DCM may include circuits to perform clock de-skew, clock phase shifting, clock frequency synthesis, and other clock features. The CCM may include circuits for phase-matched binary clock division and internal clock jitter and skew measurement.
The configuration logic includes logic used to address and load configuration information into configuration memory cells, such as SRAM-based configuration memory cells, during external configuration of FPGA 60. The configuration logic may include configuration registers, boundary scan test circuitry, such as JTAG circuitry, and encryption and/or decryption circuitry used to encrypt and/or decrypt bitstreams of configuration data loaded into and read out of FPGA 60. In FPGAs, configuration memory is used to determine programmable interconnectivity and to specify any of a variety of conditions for functional blocks, among other configuration uses. In an FPGA, there are function blocks, such as MGTs, DCMs, and a System Monitor, among other blocks, where an ability to reprogram dynamically, namely, to change these conditions in these functional blocks while the FPGA is in operational use, would be desirable.
Conventionally, the time to initially configure the configuration memory of an FPGA and to partially reconfigure the configured FPGA is relatively long compared to the operational times of using, for example, a typical decoder formed by configuring CLBs. As an illustration from p. 387 of the Virtex-II Platform FPGA Handbook, Dec. 3, 2001 from Xilinx Inc. of San Jose, Calif. If the FPGA has 404 frames at 832 bits per a frame, then with a 50 MHz configuration clock, it takes about 84 ms to initially configure the FPGA. To reconfigure one frame (832 bits) it would take about 17 usec. If the FPGA has 404 frames at 1472 bits per a frame, then with a 50 MHz configuration clock, it takes about 1.5 ms to initially configure the FPGA. To reconfigure one frame (1472 bits) it would take about 29 usec. At p. 90 of the Virtex-II Platform FPGA Handbook the internal performance of a 16-bit address decoder is about 398 MHz. Thus in one embodiment of the present invention the internal operational clock speed for the operation of a simple design such as a decoder in a configured FPGA is about an order of magnitude faster than the configuration clock speed.
The DRPORT of an exemplary embodiment of the present invention allows reconfiguration of less than a frame of configuration memory without reconfiguring the entire frame and the reconfiguration of less than a frame is done at the internal operational clock speed rather than the configuration clock speed. Thus the data rate of reconfiguring of a group of configuration memory cells less than a frame at the internal operational clock speed of a simple configured circuit is at least one order of magnitude greater than the typical reconfiguration of a frame of configuration memory at the configuration clock speed.
“Dynamic reconfiguration” as used herein means configuring or reconfiguring a set of one or more configuration memory cells at a data rate substantially greater than the traditional reconfiguration of a frame of configuration memory at the configuration clock speed. In one embodiment the operational data rate used by dynamic reconfiguration is at least an order of magnitude greater than the typical reconfiguration data rate.
To facilitate such dynamic reconfiguration, it would be desirable to have a separate configuration memory interface apart from the ICAP and convention configuration interface. What follows describes in several embodiments a dynamic reconfiguration port that may be used to dynamically reconfigure a set of configuration memory cells in an integrated circuit.
FPGA 100 includes an FPGA fabric 111, namely, a region where dedicated and programmable logic exists within an integrated circuit. FPGA fabric 111 includes a dynamic reconfiguration port (DRPORT), i.e., “reconfiguration” port 101.
Reconfiguration port 101 has access to one or more regions of an integrated circuit having programmable cells. Programmable cells may include volatile memory cells, non-volatile memory cells, and like programmable cells for reconfiguration. For clarity, programmable cells are described below in terms of memory cells, or configuration memory cells, although it will be apparent that other forms of circuitry for storing state may be used.
Reconfiguration port 101 provides access to and from function block 112 and to and from configuration logic 103 via a function block 112. Configuration logic 103 is for controlling configuration of configurable blocks of FPGA 100 and frame data registers, among other circuitry as is known. A configuration bitstream feeds into configuration logic 103 that drives frame data registers and other known circuitry. An ICAP (not shown) connects to configuration logic 103. Configurable blocks of FPGA 100 include, but not are not limited to, one or more processors 196, CLBs 197, IOBs 198 and PIPs 199, as is known. Configurable blocks of FPGA 100 may be accessed through interconnect fabric of FPGA fabric 111. Reconfiguration port 101 connects to the interconnect fabric of FPGA fabric 111 to allow access to signals of FPGA 100. Thus, for example, a user may drive reconfiguration port 101 from a user instantiated circuit design implemented in one or more CLBs 197 externally through one or more IOBs 198, or through an embedded processor 196 or a soft processor, such as may be instantiated from CLBs 197. It should be appreciated that given sufficient space for implementing, a reconfiguration port for dynamic reconfiguration may be assigned to each function block. Thus, there are multiple function blocks 112 each having a controller 102 associated therewith for reconfiguration port 101 access, and there are multiple reconfiguration ports 101 corresponding to each controller 102. As is described below in additional detail, dynamically reconfigurable memory cells are dual ported. One of these memory ports is for dynamic reconfiguration via a reconfiguration port 101. Accordingly, each reconfiguration port 101 is associated with an address space, namely, configuration memory cells, for an associated function block 112. Though shown separately, an FPGA fabric 111 reconfiguration port 101 may be provided as an integral part of each function block 112. Accordingly, any function block 112 where dynamic reconfiguration may be advantageous may have an associated reconfiguration port. Though there are multiple function blocks 112 and associated reconfiguration ports, for purposes of clarity, a single reconfiguration port 101 associated with a single function block 112 is described, as it will be apparent that multiple reconfiguration ports 101 and function blocks 112 may be used.
Signals to and from reconfiguration port 101 and controller 102 are inverted or complemented as indicated by a “_B” to denote a “bar”. Complementing may be done when connecting through an interconnect block, such as controller 102. Port interface 110 signaling from reconfiguration port 101 to controller 102 includes data clock signal 121 (“DCLK_B”), data enable signal 122 (“DEN_B”), data write enable signal 123 (“DWE_B”), data address signals 123 (“DADDR_B[m:0]”), and data input signals (“DI_B[n:0]”), where “[m:0]” indicates m to 0 address lines for m an integer greater than zero and “[n:0]” indicates n to 0 data lines for n an integer greater than zero. Accordingly, addresses of length m+1 bits may be used, and data words of length n+1 bits may be used. Notably, addresses or data may be communicated serially or in parallel, though with respect to reconfiguration port 101 parallel address and data communication bussing is generally described herein. To facilitate user adoption, an interface with similar signal timing to a BRAM interface was created though implementation of reconfiguration port 101 is substantially different than a BRAM interface.
Port interface signaling from controller 102 to reconfiguration port 101 includes data output signals 126 (“DO_B[n:0]”) and data ready signal 127 (“DRDY_B”). Data ready signal 127 is a handshaking signal provided to reconfiguration port 101 to indicate that controller 102 has completed or is about to complete the current operation and is ready for a next operation.
Reconfiguration port 101 is a read/write (“R/W”) port, where data ready signal 127 may be used for read and write wait states. Notably, in the embodiment shown, there is no read enable signal shown. This is because in this embodiment, data output signal 126 is maintained in an active state. In another embodiment, a read enable signal may be used.
Controller 102 may further be coupled to receive other signals, such as a global write enable signal and a global restore signal, among others. Such global signals may be used for added functionality, as described below in additional detail.
Controller 102 communicates with configuration logic 103 and function block logic 104 via R/W interface 105. R/W interface 105 communicates with configuration logic 103 and function block logic 104 in part to dynamically read/write configuration/reconfiguring data bits from/to configuration memory cells, such as memory cells 131 of
For this embodiment, configuration bits may be broken out into two general types, namely, bits that are reconfigurable via reconfiguration port 101 (“dynamically reconfigurable configuration bits”) and bits that are not reconfigurable via reconfiguration port 101 (“non-dynamically reconfigurable configuration bits”), where the non-dynamically reconfigurable configuration bits are reconfigured via configuration logic 103 using, for example, the ICAP or select map interface.
Configuration bit interface 109 may handle both types of configuration bits 106 though split into two sections. One of those two sections may handle only dynamically reconfigurable (DR) configuration bits 108, and thus that section or those addressable bits would be accessible by controller 102 via R/W interface 105. The other section of the two sections may handle only non-dynamically reconfigurable (NDR) bits 107, and thus would not have to be coupled to controller 102 via R/W interface 105 for communication.
More particularly, R/W interface 105 is coupled to memory cells 131, as described with reference to
Notably, in
Thus, it should be appreciated that all of memory cells 130 may be configured or reconfigured though conventional non-dynamic means. However, with respect to use of reconfiguration port 101, only a portion of memory cells 130, namely, memory cells 131, are accessible via controller 102 for dynamic reconfiguration. As mentioned above, memory cells 130 may be any of a variety of known types of memory cells operable at relatively high frequencies, and in an embodiment is equal to or in excess of approximately 500 MHz. In one embodiment, the approximately 500 MHz, for example, is a frequency of operation with no wait states (i.e., either one read or one write per data clock cycle) for use of function block 112.
For clarity by way of example, memory cells 131 are described as having SRAM memory elements, though other known types of memory elements may be used. Memory cells 131 are dual ported, where one port is for conventional non-dynamic access for configuration data and address information, and the other port is for dynamic read/write access under the control of controller 102. The term “port” as used herein includes one or more signal pins or nodes, and may refer to a particular signal pin or node.
Notably, controller 102 may be configured to access all memory elements of memory cells 131 or only a subset thereof which may depend on functionality of function block 112.
Returning to
Notably, function block 112 has not been described in terms of a particular type of function, as any of a variety of functions may be used. Some examples of functions that may be used for function block 112 include without limitation a Digital Signal Processor (“DSP”), an MGT, a DCM, a CLB, an IOB, or a System Monitor. Furthermore, controller 102 may include additional functionality for one or more of such function blocks. Moreover, interconnect switch matrices and the like including one or more PIPs may be configured via a reconfiguration port 101.
For example, depending on function block logic 104, controller 102 may optionally include function enable(s) interface 114 and block status interface 113. Block status interface 113 may be a read-only port by controller 102, and function enable(s) interface 114 may be a write-only port by controller 102. Interfaces 113 and 114 may be in addition to R/W interface 105.
Block state may be read from status signaling from function block logic, and a function may be activated by providing one or more function enables to function block logic 104. Block status signaling and function enable signaling may be accessed from reconfiguration port 101 via controller 102 by addressing.
For example, address space may be broken out into distinct groups for addressable memory and addressable signals. Responsive to receiving an address within an address space assigned to addressable memory, such as via data address signal 124, controller 102 accesses R/W interface 105 to write to or read from a memory element of a memory cell, or respective memory elements of memory cells. Responsive to receiving an address within an address space assigned to a signals/functions (including without limitation a block status request) controller 102 writes a function enable via interface 113 into the block or read status output. This disables reading/writing from one or more memory cells, and causes status signaling to be read back via interface 114 or causes function enables to be written into registers or other blocks in function block 104. Notably, by increasing internal address space, functions, including without limitation test functions, may be initiated with controller 102 and status read out, including without limitation test results, using reconfiguration port 101 without the expense of dedicated function and read out ports.
On rising edge 201, data enable signal 222 at 202 and data write enable signal 223 at 203 are both active high causing both data address signal 224 and data input signal 225 to be sampled (e.g., “bb” and “BB”, respectively) at approximately the same time. Controller 102 of
On rising edge 211, data enable signal 222 at 212 is active high causing data address signal 224 to be sampled (e.g., “aa”). Controller 102 of
With continuing reference to
Dual ported memory cell 300 includes storage element 301 for storage of a dynamically reconfigurable configuration bit. Notably, storage element 301 may be a memory storage element, a volatile or non-volatile storage elementor any other element capable of storing a bit state. For purposes of clarity, a memory storage element is described herein.
Data signal (“D”) 304 is connected to a source terminal of pass transistor 321. An inverted version of data signal 304, namely, data signal (“D_B”) 303, is connected to a source terminal of pass transistor 322. Drain terminals of transistors 321 and 322 are respectively conventionally coupled to a cross-coupled latch of a conventional SRAM memory element. An address signal (“A”) 302 is applied to gates of pass transistors 321 and 322 for selectively coupling data signal 304 to an input of an inverter formed by p-type transistor 325 and n-type transistor 326, and for selectively coupling data signal 303 to an inverter formed by p-type transistor 323 and n-type transistor 324. These inverters are cross-coupled to form a latch. Sources of transistors 323 and 325 are coupled to a supply voltage, such as VDD 327, and sources of transistors 324 and 326 are coupled to a ground potential, such as ground 328. Output from the inverter formed of transistors 323 and 324 is output signal (“Q”) 306. Output from the inverter formed of transistors 325 and 326 is output signal (“Q_B”) 305.
It should be understood that ports for data signals 303 and 304 and output signals 305 and 306 are bi-directional, namely, they are input/output ports. Furthermore, it should be understood that pass transistors 321 and 322 provide one port of dual ported memory cell 300, and the other port is provided via pass transistors 310 and 311. As memory element 301 is conventional, further description is omitted for purposes of clarity. R/W interface 105 of
For access via reconfiguration memory cell port 330 for a write operation, a write address signal (“W_A”) 309 is provided to each gate of pass transistors 310 and 311. A source terminal of pass transistor 310 sources an inverted write port signal (“W_P_B”) 307, and a source terminal of pass transistor 311 sources a write port signal (“W_P”) 308. Drain terminals of pass transistors 310 and 311 are respectively coupled to signal paths of output signals 305 and 306 for input to memory element 301.
Responsive to write address signal 309 being at a high logic level, transistor 310 couples inverted write port signal 307 to inverted output signal 305 as an input to memory element 301, and transistor 311 couples write port signal 307 to inverted output signal 305 as an input to memory element 301 for writing any data on write port signals 307 and 308 to memory element 301. Thus, responsive to write address signal 309 being asserted, write port signals 307, 308 cause a dynamically reconfigurable configuration bit to be written to memory element 301.
During a read operation via reconfiguration memory cell port 330, write address signal 309 is at a low logic level. A state stored in a latch of memory cell 301 is output via output signal paths for output signal 305, 306. By connecting to one of these paths, data may be read from memory element 301.
Data read from memory element 301 for one of the ports is sourced from output signal 306 and for reconfiguration memory cell port 330 is sourced from output signal 305. Output from output signal 305 is provided as an input to inverter 332, and output from inverter 332 is provided as an input to select transistors 314 and 315. Output from output signal 305 may optionally be provided as an input to inverter 363, where output from inverter 363 is additional output signal, namely, output signal (“Q2”) 331.
Because memory element 301 is continuously read, except during a write operation, select circuitry, such as with n-type transistor 314 and p-type transistor 315 coupled in parallel, may be used to controllably select when data is to be readout as read port signal (“R_P”) 335. Notably, other select circuits may be used, such as a single pass transistor coupled in series with inverter 332 to provide read port signal (“R_P”) 335, among other known types of select circuits.
Transistors 314 and 315 are sourced with the output from inverter 332, and transistors 314 and 315 have their drains commonly connected to a read port node 336. Transistor 314 is gated with read address signal (“R_A”) 333, and transistor 315 is gated with an inverted version of read address signal 333, namely, inverted read address signal (“R_A_B”) 334. Responsive to read address signal 333 being at a high logic level and read address signal 334 being at a low logic level, both of transistors 314 and 315 conduct output of inverter 332 to read port node 336 to provide read port signal 335. Responsive to read address signal 333 being at a low logic level and read address signal 334 being at a high logic level, both of transistors 314 and 315 do not conduct output of inverter 332 to read port node 336.
Notably, read port signal 336 may be provided to a multiplexer, and write address 309 and read address 334 may be obtained from a decoder.
Moreover, it may be desirable to do a chip-wide check of memory cells used for non-dynamically reconfigurable bits for an inadvertent change of state, such as may be caused by a Single Event Upset (“SEU”) due to subatomic particles. However, memory cells used for dynamically reconfigurable configuration bits may be masked from such an SEU check, as dynamically reconfigurable configuration bits may frequently be intentionally changed. However, a user may desire to opt out of dynamic reconfiguration, in which event it may be desirable to be able to selectively enable and disable masking for SEU checking of memory cells allocated for dynamic reconfiguration. Furthermore, within a frame, a user may desire to frequently change only a fraction of the bits of the frame, in which event it may be desirable to be able to selectively enable masking out SEU checking for those bits to be frequently changed in blocks of memory cells, while allowing SEU checking for the remainder of the bits of in the frame. Masking of dynamically reconfigurable configuration bits is described below in additional detail.
Though write and read ports of memory cell 300 were described in terms of storing a dynamically reconfigurable configuration bit, as mentioned above write-only and read-only ports may be provided for function enable(s) and block status, respectively. Thus, a portion of addressable memory cells 300 may be reserved as dynamically reconfigurable memory cells, and another portion of addressable memory cells 300 may be reserved to provide write-only and read-only ports by connecting write and read ports of such memory cells 300, not to memory elements 301, but to status registers.
Memory cell frame architecture 400 includes upper frame configuration bit section 401 and lower frame configuration bit section 402. Disposed between upper and lower frame sections 401 and 402 is a block of memory cells 404 for routing global signals with respect to frame 400, which may for example be used for masking, as described in additional detail in a co-pending patent application entitled “DATA MONITORING FOR SINGLE EVENT UPSET IN A PROGRAMMABLE LOGIC DEVICE” by Martin L. Voogel, et al., filed Mar. 22, 2004, which is incorporated by reference herein in its entirety.
Upper and lower frame sections 401 and 402 each include N blocks 403 of configuration memory cells, such as dual ported memory cells 300 of
For a block 403 having 20 memory cells, 16 memory cells 413 may be used for storing dynamically reconfigurable configuration bits. To provide address separation or granularity, memory cells 410-1 and 410-2 have their address line electrically open (“broken”) for address signal 302 and are partially unconnected or disconnected from memory cells 413. Memory cell 412 may be used to store a masking bit, and memory cell 411 may be left unused. Masking circuitry for a masking bit is described in additional detail in above-mentioned co-pending patent application entitled “DATA MONITORING FOR SINGLE EVENT UPSET IN A PROGRAMMABLE LOGIC DEVICE” by Martin L. Voogel, et al., which was incorporated by reference herein in its entirety. A masking bit may be used herein to prevent SEU checking of one or more memory cells in a block that stores dynamically reconfigurable configuration bit information. In the above-incorporated reference, 12 memory cells of 16 memory cells of block 404 are reserved to provide a checksum or like check for SEU checking. Thus, the remaining 4 memory cells may be used for another purpose like the remainder of block 403.
It should be understood that memory cells of frame 400 are configuration memory cells for configuring logic, a portion of which may be reserved for dynamically reconfigurable configuration bits for interaction with function block logic 104 of
Masking bits are applied locally to a group of memory cells, such as memory cells 413 of block 403, but may depend in part upon application of the block. Accordingly, for example, masking may be done with a 16-bit granularity. Notably, masking may be at a higher level (e.g., at a frame level) instead of locally at a block level, or done at a level between a block level and a frame level depending on application.
Because memory cells, such as memory cells 300 of
Again though configuration memory cells have been described in terms of an FPGA, it should be appreciated that other types of integrated circuits may be used as previously mentioned, and that cells, other than memory cells, may be used as previously mentioned and such cells may be connected to signals instead of memory elements. Moreover, though a block of 16 memory cells was described for storing dynamically reconfigurable configuration bits, it should be understood that fewer or more than 16 bits may be address separated. For example, as few as one bit may be address separated, or as many as all bits in a frame may be combined. However, with respect to providing a fine granularity for accessing one or more dynamically reconfigurable configuration bits, it should be appreciated that reconfiguration is limited to the isolated bit or bits, and not adjacent bits. With such granularity, such isolated bit or bits need not be reset during reconfiguration, though they may be reset. Moreover, with such granularity, one or only a relatively small number of bits as compared to an entire frame of bits may be changed during dynamic reconfiguration, which facilitates changing functionality while a function block is still operational. In other words, memory cells in an FPGA may be coded into software attributes for users using FPGA manufacturer provided software, for example which have heretofore generally been thought of as hard coded, approach the flexibility of off-chip input signals.
Referring to
Memory cells 410-1, 410-2 are coupled to receive address signal 302 as a address signal input. Address local signal 502 from masking circuit 700 is provided to memory cells 410-1, 410-2 and memory cells 413 as an address signal. Address signal 302 is provided to memory cells 411 and 412 as an address signal.
Memory cell 412 is coupled to receive data signal 503 and its complement data signal 504 for bit 18 (“D<18>” and “D_B<18>”) of block 403. Memory cells 413 are coupled to receive, in parallel though not specifically shown with separate lines for purposes of clarity, data signals 506 (“D<16:1>”) and their respective complemented data signals 507 (“D_B<16:1>”). Data signals 503 and 506 are related to data signal 304 of
Output signals (“Q<16:1>”) 516 and their respective complemented output signals (“Q_B<16:1>”) 517 are respectively related to output signals 306 and 305 of
Write address signal 509 and read address signals 533, 534 are provided from coordinate-to-address converter 510 to memory cells 413. Coordinate-to-address converter 510 may be considered part of or separate from controller 102 of
Write enable signal (“WE”) 523 is provided to coordinate-to-address converter 510. Responsive to write enable signal 523 being at a logic low level, coordinate signals 521, 522 are for read addressing. Responsive to write enable signal 523 being at a logic high level, coordinate signals 521, 522 are for write addressing.
Masking circuit 700 receives address signal 302, complemented output signal 501 and a complemented global masking signal (“GMASK_B”) 505, and responsive to such inputs provides address location signal 502.
Notably, data address signals 910 and 911 may form data address signal 224 of
Returning to
Coordinate signals 521, 522 are provided to NAND gate 602. The output of NAND gate 602 is coupled to inverter 605 via inverter 604. Inverters 604 and 605 are connected in series. The output of inverter 604 is tapped at node 606 to obtain read address signal 533, and the output of inverter 605 provides inverted read address signal 534.
In this exemplary embodiment, data ready signal (“DRDY”) 1014 is generated from complemented data clock signal 121 and data enable data ready (“DEN_DRDY”) signal 1015 provided to data ready signal generator 1000. Data ready signal generator 1000 includes flip-flops 1002 and 1003 connected in series and inverter 1004. Complemented data clock signal 121 is provided to an input of inverter 1004 to provide a data clock signal 1021 for clocking flip-flops 1002 and 1003.
Data enable data ready signal 1015 is provided to a data input port of flip-flop 1002, which in response to data enable data ready signal 1015 and a data clock signal 1021 input, provides an output to a data input of flip-flop 1003. Flip-flop 1003 in response to receiving output data from flip-flop 1002 and a data clock signal 1021 input provides an output which is data ready signal 1014. Notably, having two flip-flops in series, namely, two stages of flip-flops, ensures that data ready signal 1014 will not be earlier than at least two clock cycles of data clock signal 121. Data enable data ready signal 1015, may be a pipelined signal, as described below in additional detail, thereby adding at least one other clock cycle prior to indicating that controller 102 is ready for a next operation. In other words, data ready signal 1014 may be produced from at least three stages of flip-flops, which ensures at least three clock cycles of data clock signal 1021 transpire prior to indicating controller 102 is ready for a next operation. However, as illustratively shown in
Notably, signaling for controller 102 may be pipelined at or about the operating frequency of integrated circuit 100. During a read operation, output is driven by reading a memory element, which may take some time to be read out. With respect to write operations, they may be done within a single clock cycle as memory elements are relatively easy to flip. Furthermore, because a read-modify-write sequence to one or more memory cells may be used, a single memory cell may be changed at a time by reading 16 memory cells and writing the same value back into all but one of the memory cells.
Because a relatively small portion of memory cells may be used for dynamically reconfigurable configuration bits, FPGA 100 may be dynamically reconfigured using such portion of memory cells at speeds proximal or equal to frequency of operation of FPGA 100. Additionally, providing a data ready signal allows for faster data clock speeds interfacing to different blocks each of which return a data ready signal subject to their respective operating parameters.
Controller logic block 1001 receives or outputs complemented data address signal 124, complemented data enable signal 122, complemented data output signal 126, complemented data input signal 125, complemented data write enable signal 123, complemented data ready signal 127, complemented data clock signal 121, write enable signal 523, address signal 910, data ready signal 1014, write port signal 508, complemented write port signal 524, and read port signal 535, all of which have previously been described. Additionally, controller logic block 1001 receives complemented configuration reset signal (“CFG_RESET_B”) 1013, complemented global write enable signal (“GWE_B”) 1011 and complemented global restore signal (“GRESTORE_B”) 1012. In addition to use for controller logic block 1001, these three global signals 1011, 1012 and 1013 act as a chip-wide read or write enable.
Global write enable signal 1011 may be invoked to disable dynamic writing to all memory elements of FPGA 100 to allow for a conventional configuration of FPGA 100, whether externally though a configuration bit interface or internally through an ICAP. As mentioned above, memory elements may be coupled to registers, such as flip-flops, to store bit values. To write such stored bit values back to such registers, global restore signal 1012 may be invoked. Global restore signal 1012 may be provided to flip-flops 1002 and 1003 as a reset signal input. Configuration reset signal 1013 may be used to reset registers or flip-flops when the entire FPGA is being reset.
Controller logic block 1001 outputs data write enable signal 1016 and data enable signal 1017. Data write enable signal 1016 and data enable signal 1017 may be pipelined. Furthermore, address or data address signal 910 and data enable ready signal 1015 may be pipelined.
Notably, different function logic blocks may use different means for obtaining a pipelined data ready signal.
Inputs to NOR gate 1057 are output of flip-flop 1003 and function block logic data ready signals (“FBL-DRDY”) 1014A and 1014B. Signal 1014A is a first (“<0>”) bit and signal 1014B is a second (“<1>”) bit of function block logic data ready signals. Output from NOR gate 1057 is provided to inverter 1058, and the output of inverter 1058 is data ready signal 1014. Thus, function block logic control signals, such as signals 1014A and 1014B, may be used to provide data ready signal 1014.
With reference to
With simultaneous reference to
With reference to
With reference to
With reference to
If a status signal output, rather than a read memory cell output is to be obtained, memory enable is disabled. This may be done by toggling a data enable data ready signal 1015 responsive to whether either a status signal or a read memory state is to be output. Data enable signal 1017 is provided as an input to NAND gate 1145. Another input to NAND gate 1145 is an address bit to indicate whether either status or memory state is to be output. Output of NAND gate 1145 is provided as an input to inverter 1146. The output of inverter 1146 is data enable data ready signal 1015.
For purposes of clarity by way of example and not limitation, a seventh bit of address signal 910 (recall it was assumed that address signal was seven bits <6:0>) may be used, namely, pipelined address signal (“ADDRESS<6>”) 910-6, as an input to NAND gate 1145. Of course, other addressing schemes may be used to toggle between status and memory states.
Assume that if a bit value of address signal 910-6 is a logic zero, that a status signal is to be read by controller 102 of
Inputs to NAND gate 1143 are pipelined data write enable signal 1016, pipelined address signal 910-6, pipelined data enable signal 1017, and complemented data clock signal 121. The output of NAND gate 1143 is provided as an input to inverter 1144. The output of inverter 1144 is write enable signal 523.
All inputs to NAND gate 1143 are pipelined, except for complemented data clock signal 1021, and thus they will be approximately synchronous with one another. Moreover, as flip-flops 1132, 1141, and 1142 are clocked off of data clock signal 1021, complemented data clock signal 121 will be approximately synchronous with pipelined data write enable signal 1016, pipelined address signal 910-6, and pipelined data enable signal 1017.
Accordingly, when data write enable signal 1016, address signal 910-6, and data enable signal 1017 are all logic one, output of NAND gate will toggle with complemented data clock signal 121 though it will be the inverse value. For example, if data write enable signal 1016, address signal 910-6, and data enable signal 1017 are all logic one and complemented data clock signal is logic one, output of NAND gate 1143 will be a logic zero. For example, if data write enable signal 1016, address signal 910-6, and data enable signal 1017 are all logic one and complemented data clock signal is logic zero, output of NAND gate 1143 will be a logic one. Thus, if data write enable signal 1016, address signal 910-6, and data enable signal 1017 are all logic one, by obtaining write enable signal 523 from the output of inverter 1144, write enable signal will be a logic one when complemented data clock signal 121 is a logic one. Moreover, if data write enable signal 1016, address signal 910-6, and data enable signal 1017 are all logic one, by obtaining write enable signal 523 from the output of inverter 1144, write enable signal will be a logic zero when complemented data clock signal 121 is a logic zero.
If any of data write enable signal 1016, address signal 910-6, and data enable signal 1017 are a logic zero, output of NAND gate 1143 will be a logic one. Thus, output of inverter 1144 will be a logic zero indicating that writing is not enabled, as write enable signal 523 will be a logic zero.
Accordingly, it should be appreciated that by using clocked latches, such as flip-flops, for pipelining to obtain write port signals 508 and 524, data write enable signal 1016, address signal 910, and data enable signal 1017, and thus obtain write enable signal 523 as described above, write access of memory may be done at the same frequency of operation of the integrated circuit in which the memory is disposed.
With reference to
Prior to transfer over to the pipeline, data enable signal 222 and data write enable signal 223 may be logic one state, and data address valid signal (“DADDR<6>”) 124-6 is valid. Signals 222, 223 and 124-6 may be clocked by an external clock signal (not shown) which may be the source of data clock signal 1021.
Responsive to rising edge 1201, data enable pipeline signal 1017 and data write enable pipeline signal 1016 transition at 1205 from a don't care condition to a logic one state. Responsive to rising edge 1201, data address pipeline signal 910-6 transitions at 1207 from a don't care condition to a valid address output. Accordingly, responsive to transfer to pipelining, data enable signal 222 and data write enable signal 223 transition at 1204 to a don't care condition, and data address signal 124-6 transitions at 1206 to a don't care condition. Thus, data of data address signal 124-6 is now pipeline data of data address signal 910-6.
Write enable signal 523 goes from an off logic low state to an on logic high state responsive to falling edge 1202 of data clock signal 1021. In other words, rising edge 1212 of write enable signal 523 is responsive to falling edge 1202 of data clock signal 1021. Write enable signal 523 is held in a logic high state until a next rising edge 1203 of data clock signal 1021. In other words, falling edge 1213 of write enable signal 523 is responsive to rising edge 1203 of data clock signal 1021. Thus, write enable signal 523 is active for approximately one-half of a clock cycle of data clock signal 1021. Responsive to such next rising edge 1203 of data clock signal 1021, pipeline signals 1016, 1017 and 910-6 transition to don't care conditions.
Controller 102 and decoder 900 may be at least approximately centrally located with respect to memory cells 1300. Notably, decoder 900 may be considered part of controller 102. With respect to address, including but not limited to row address, signaling and global masking signaling, they may be done as previously described, though for this memory array architecture. For example, an address signal A<5:0> may be used to address blocks 1301–1306 where A<0> for example is used to address a particular block. One unique y-bit out of column signal 922 of
Additional, digital clock manager logic may be used to interact with above-described data address and data ready signaling. Read data and data ready signals may be ready two clock cycles after a data enable signal.
Controller 102 and decoder 900 may be at least approximately centrally located with respect to memory cells 1300. A bit of address signal 910 of
One unique y-bit out of column address signal 922 is routed to each block 1401–1408 for block addressing. Read data and data ready signals may be ready two clock cycles after a data enable signal.
Registers 1603 and 1604, accessed via reconfiguration port 101 through controller 102, may be initially set with default settings obtained from a configuration bitstream. Thus, system monitor 1600 may start in a known state. Additionally, alarm values to be stored in memory cells 1500 may be downloaded from FPGA 100 configuration memory.
Registers 1603 include system monitor configuration registers, test registers and channel sequence registers. Registers 1604 are used to hold alarm thresholds for a digital comparison and calibration coefficients for on-chip sensors.
Additional details regarding system monitor 1600 may be found in co-pending, concurrently filed patent application entitled “DYNAMIC RECONFIGURATION OF A SYSTEM MONITOR (DRPORT)” by F. Erich Goetting, et. al., which is incorporated herein by reference in its entirety.
While the foregoing describes exemplary embodiment(s) in accordance with one or more aspects of the invention, other and further embodiment(s) in accordance with the one or more aspects of the invention may be devised without departing from the scope thereof, which is determined by the claim(s) that follow and equivalents thereof. It should be appreciated that the above-described dynamic reconfiguration port includes a memory interface that appears like a well-known memory interface, in particular a memory interface for a BRAM of an FPGA. This facilitates compatibility with microprocessors/microcontrollers, whether formed of dedicated circuitry or configurable logic or any combination thereof. For example, a memory block may be mapped into memory or input/output space of a microprocessor. Because mapping a memory block into such space is a well-known model for on-chip control of microprocessor controlled peripherals, the above-described interface is compatible with well-known on-chip interfaces of interconnecting cores, such as a CoreConnect from IBM or AMBA from ARM.
Claim(s) listing steps do not imply any order of the steps. Trademarks are the property of their respective owners.
Number | Name | Date | Kind |
---|---|---|---|
5586299 | Wakerly | Dec 1996 | A |
5774474 | Narayanan et al. | Jun 1998 | A |
5999480 | Ong et al. | Dec 1999 | A |
6034542 | Ridgeway | Mar 2000 | A |
6191614 | Schultz et al. | Feb 2001 | B1 |
6201728 | Narui et al. | Mar 2001 | B1 |
6204687 | Schultz et al. | Mar 2001 | B1 |
6255848 | Schultz et al. | Jul 2001 | B1 |
6262596 | Schultz et al. | Jul 2001 | B1 |
6429682 | Schultz et al. | Aug 2002 | B1 |
6732354 | Ebeling et al. | May 2004 | B1 |
6836839 | Master et al. | Dec 2004 | B1 |
6851047 | Fox et al. | Feb 2005 | B1 |
20020138716 | Master et al. | Sep 2002 | A1 |
20030034848 | Norman et al. | Feb 2003 | A1 |
20030105949 | Master et al. | Jun 2003 | A1 |
20030154357 | Master et al. | Aug 2003 | A1 |
20040030736 | Scheuermann | Feb 2004 | A1 |
20040078403 | Scheuermann et al. | Apr 2004 | A1 |
20050044344 | Stevens | Feb 2005 | A1 |
Number | Date | Country |
---|---|---|
0 748 051 | Dec 1996 | EP |
Number | Date | Country | |
---|---|---|---|
20050242835 A1 | Nov 2005 | US |