Method and Apparatus for Digitally Compensating Digital Clock Skew for High Speed Digital Circuits

1. Field of the Invention

The present invention relates to the field of high speed digital circuits, in particular, high speed digital circuits based on CMOS technology. More specifically, the present invention relates to a method and apparatus for digitally compensating digital clock skew for high speed digital circuits in a digital system, such as a data instrumentation system.

2. Background

In a digital system having a number of digital circuits, such as a data instrumentation system, as the digital clock passes through one of the digital circuits, the clock high time tends to either shrink or expand, resulting in the skewing of the digital clock. In some CMOS circuits, the skewing of the digital clock, i.e. variations in the clock high time and clock low time, can be more than two (2) or three (3) nanoseconds (ns) depending on the loading of the particular CMOS circuits. A 2-3 ns skew can be significant, depending on the length of the digital clock period. Experience has shown that 2-3 ns skew actually become limiting to either the speed of the system or the complexity of the logic as the digital clock period approaches the range of 10-15 ns. The digital clock period approaches the range of 10-15 ns, as the operating speed approaches about 66 MHZ. As a result, 66 MHZ has become the de facto maximum operating speed for many digital designs based on CMOS high speed circuits under the current 0.8 micron technology. Since it is obviously beneficial to be able to improve and push the maximum operating speed of digital designs based on CMOS high speed circuits beyond 66 MHZ, it is desirable to be able to digitally compensate the digital clock skew as it travels through one CMOS high speed circuit onto another.

Additionally, for a digital system such as a data instrumentation system, it is often operated at different speeds at different times. However, depending on the design and the load, the CMOS high speed circuits in the digital system require a certain minimum amount of high and low time in each clock period in order to function properly regardless of the operating frequency. Thus, it is further desirable to be able to ensure that the minimum amount of high and low time are provided when digitally compensating digital clock skew.

As will be disclosed, the present invention provides such a method and apparatus which advantageously achieves the desirable results described above. As will be obvious from the descriptions to follow, the present invention has particular application to high speed data instrumentation systems.

Under the present invention, the desirable results are advantageously achieved by digitally reconstructing the digital clock as it enters a high speed digital circuit after traveling through another high speed digital circuit. Each high speed digital circuit is provided with a digital clock reconstruction circuit for reconstructing the entering digital clock. The digital clock reconstruction circuit comprises a flip flop, a programmable delay chain, and an assembly of gates, coupled to each other in a circular manner. The flip flop generates the reconstructed digital clock using the entering clock and either a regular or a shortened clear signal provided by the assembly of gates. The reconstructed digital clock and its complement are provided to the programmable delay chain, which in turn uses them to generate input signals for the assembly of gates for the generation of the regular or shortened clear signal for the flip flop.

Additionally, each digital high speed circuit is provided with one or more measurement or comparison circuits for measuring the frequencies of various reconstructed clocks generated under different delay settings of the programmable delay chain or comparing these frequencies to the frequency of the entering clock, to determine the delay setting for the programmable delay chain for achieving frequency symmetry between the entering clock and the reconstructed clock, i.e. same number of clocks, and 50% duty cycle symmetry for the reconstructed clock period, i.e. an equal amount of high and low time, thereby calibrating the digital clock reconstruction circuit.

The digital clock reconstruction circuit is calibrated using the maximum operating frequency. The proper delay setting for the programmable delay chain is determined iteratively, starting with an initial delay setting and varying the delay setting systematically in a predetermined manner. As a result, when frequency symmetry is achieved between the reconstructed and the entering clocks, 50% duty cycle symmetry for the reconstructed clock period is also simultaneously achieved.

Preferably, the digital high speed circuit is also provided with a ring oscillator comprising a number of rings, each having a known number of delay elements used in the programmable delay chain, for generating different ring signals. The frequencies of the ring signals, determined using one of the measurement circuits, are used to infer the operational delay yield of a delay element in the programmable delay chain, thereby providing guidance for the selection of the calibration starting delay setting for the programmable delay chain.

Furthermore, it is preferred if at least one of the measurement circuits provided can actually measure the high and low time of a reconstructed clock period during calibration, and at least one of the measurement or comparison circuits provided is used to periodically monitor and determine whether the reconstructed and entering clocks remain symmetrical to each other in frequency during normal operation. In fact, it is further preferred that the digital clock reconstruction circuit comprises an additional assembly of flip flop and gate coupled to the programmable delay chain in a similar manner. The additional flip flop is used to generate an additional reconstructed clock using the entering clock and an additional regular clear signal provided by the additional gate assembly. The additional reconstructed clock, generated using the additional regular clear signal, is periodically monitored for providing early warning to the fact that the reconstructed clock period is beginning to drift away from the 50% duty cycle symmetry.

The objects, features, and advantages of the present invention will be apparent from the following detailed description of the presently preferred and alternate embodiments of the invention with references to the drawings in which:

FIG. 1 illustrates an exemplary data instrumentation system incorporated with the teachings of the present invention.

FIG. 2 illustrates the trace board of FIG. 1 in further detail.

FIG. 3 illustrates the digital clock reconstruction and related circuitry of the present invention in each of the TARs of FIG. 2.

FIG. 4 illustrates the presently preferred embodiment of the digital clock reconstruction circuit of FIG. 3.

FIGS. 5a-5d illustrate the timing of the various signals for reconstructing the skewed entering digital clock.

FIGS. 6a-6e illustrate one embodiment of the programmable delay chain of FIG. 4 in further detail.

FIGS. 7a-7b illustrate one embodiment of the first and second gate assemblies of FIG. 4 in further detail.

FIG. 8 illustrates one embodiment of a counter-based frequency measurement circuit of one of the measurement/comparison circuits of FIG. 3 in further detail.

FIG. 9 illustrates one embodiment of the ring oscillator of FIG. 3 in further detail.

FIG. 10 illustrates the method of the present invention.

In the following description for purposes of explanation, specific numbers, materials and configurations are set forth in order to provide a thorough understanding of the present invention. However, it will be apparent to one skilled in the art that the present invention may be practiced without the specific details. In other instances, well known systems are shown in diagrammatical or block diagram form in order not to obscure the present invention unnecessarily.

Referring now to FIG. 1, a block diagram illustrating an exemplary data instrumentation system incorporated with the teachings of the present invention is shown. Shown is an exemplary data instrumentation system 10 incorporated with the teachings of the present invention coupled to a target system 11 for monitoring, acquiring data, and controlling the target system 11. Depending on the target system 11, the data instrumentation system 10 may be operated with different operating speeds at different times. The exemplary data instrumentation system 10 comprises a processor board 12, a trace board 14 incorporated with the teachings of the present invention, and a number of probes 16, while the target system 11 comprises an emulation processor. The processor board 12 is coupled to the trace board 14, for example, through a parallel bus 20. The trace board 14 is coupled to the probes 16, for example, through a number of parallel cables 22. The probes 16 are in turn coupled to the emulation processor 18. The trace board 14 will be described below in further detail with additional references to the remaining figures. The processor board 12, the probes 16 and the emulation processor 18 are intended to represent a broad category of these elements found in many data instrumentation and target systems. Their constitutions and functions are well known and will not be further described.

Referring now to FIG. 2, a block diagram illustrating the trace board of FIG. 1 in further detail is shown. The trace board 14 comprises a plurality of high speed trace and recognition circuitry (TAR) 30a-30d incorporated with the teachings of the present invention, a number of probe connectors 26a-26d, a bus controller 34, and trace control circuitry 32. The probe connectors 26a-26d, the bus controller 34, and the trace control circuitry 32 are all coupled to the high speed TARs 30a-30d. The probe connectors 26a-26d are also coupled to the bus controller 34 and the trace control circuitry 32. Each of the high speed TARs 30a-30d receives probe data along with its own digital clock from a probe 16a-16d through its corresponding probe connector 26a-26d. The digital clocks are skewed as they travel from the probes 16a-16d to the TARs 30a-30d. While each of the high speed TARs 30a-30d may operate at different operating speeds for different target systems, its components do require a minimum amount of high and low time in each clock period of the digital clock it receives along with the probe data. Thus, each high speed TAR 30a-30d is provided with digital clock reconstruction and related circuitry of the present invention for compensating the entering clock's skew, and ensuring the required amount of high and low time is provided. The digital clock reconstruction and related circuitry provided to each of the high speed TARs 30a-30d will be described below in further detail with additional references to the remaining figures. Otherwise, the high speed TARs 30a-30d, the probe connectors 26a-26d, the bus controller 34, and the trace control circuitry 32 are intended to represent a broad category of these elements found in many trace boards, including but not limited the trace boards described in copending U.S. patent applications, Ser. No. 08/040,902, entitled Method and Apparatus For Deskewing/Resychronizing Data Slices With Variable Skews, and Ser. No. 08/040,901, entitled Method and Apparatus For Synchronizing Periodic Sync Pulse Generations By A Number Of High Speed Circuits, both assigned to assignee of the present invention, which are hereby fully incorporated by reference. Their constitutions and functions will not be further described.

While the present invention is being described with high speed TARs of a trace board of a data instrumentation system, based on the description to follow, it will be appreciated that the present invention may be practiced with other high speed digital circuits on the data instrumentation system. In fact, the present invention may be practiced with other digital systems comprising high speed circuit, such as a high speed microprocessor based computer system.

Referring now to FIG. 3, a block diagram illustrating the digital clock reconstruction and related circuitry of the present invention provided to each of the TARs of FIG. 2 is shown. As illustrated, each TAR is provided with a digital clock reconstruction circuit 36 and at a minimum, a measurement or a comparison circuit 44. Preferably, for improved reliability, multiple measurement or comparison circuits 44 are provided, and for improved useability, a ring oscillator 64 is also provided. The digital clock reconstruction circuit 36 is used to generate a first reconstructed clock (CLK.sub.-- RECNSTR), and preferably a second reconstructed clock (CLK.sub.-- RECNSTR') based on the entering clock (CLK.sub.-- IN) and a number of implementation dependent control signals (CONTROLS) provided by the trace control circuitry 32. When calibrated, CLK.sub.-- RECNSTR corrects the digital skew of CLK.sub.-- IN and meets the minimum high and low time requirement of the TAR regardless of the ultimate operating speed. CLK.sub.-- RECNSTR' is used as a basis during normal operation for providing early warning to the fact that the clock period of CLK.sub.-- RECNSTR is starting to drift away from 50% duty cycle symmetry. The measurement or comparison circuits 44 are used to selectively measure the frequencies of CLK.sub.-- RECNSTR and CLK.sub.-- RECNSTR' or compare them to the frequency of CLK.sub.-- IN during calibration and normal operation of the digital clock reconstruction circuit 36. The ring oscillator 64 is used to generate a number of ring signals during calibration. The digital clock reconstruction circuit 36, the measurement and comparison circuits 44, the ring oscillator 64, the difference between CLK.sub.-- RECNSTR and CLK.sub.-- RECNSTR', the calibration process, and monitoring during normal operation, will be described in further detail below with references to the remaining figures.

Referring now to FIGS. 4 and 5a-5d, a block diagram and four timing diagrams illustrating the preferred embodiment of the digital clock reconstruction of the present invention of FIG. 3 are shown. As illustrated in FIG. 4, the digital clock reconstruction circuit 36 comprises a first flip flop 38a, a programmable delay chain 40, a first assembly of gates 42a, and preferably a second flip flop 38b, and a second assembly of gates 42b. The first flip flop 38a, the programmable delay chain 40, and the first assembly of gates 42a are serially coupled to each other in a circular manner. The programmable delay chain 40 is also coupled to the second assembly of gates 42b, which in turn is coupled to the second flip flop 38b. For this embodiment, the implementation dependent control signals (CONTROLS) provided by the trace control circuitry 32 include a constant high enable (CLKCLR.sub.-- N), a delay setting selection signal (TAPSEL), and an early clear enable signal (ECLR.sub.-- N).

The first flip flop 38a is used to ,generate CLK.sub.-- RECNSTR and its complement. The first flip flop 38a is a type D rising clock edge triggered flip flop. The first flip flop 38a receives a constant high enable as its D input, and the entering digital clock as its clock input. At each clock period, with active low not asserted at the clear input, the constant high enable at the D input is clocked into the first flip flop 38a at the rising clock edge of the clock pulse. Concurrently, a logic high and a logic low are clocked out of the first flip flop 38a at the Q and Q' outputs. The Q and Q' outputs remain at high and low respectively, until an active low is asserted at the clear input of the first flip flop 38a, clearing its content and inverting the Q and Q' outputs. At that time, the content of the first flip flop 38a changes to logic low, and the Q and Q' outputs change to logic low and high respectively. The Q and Q' outputs remain at low and high respectively, until the active low is deasserted. As this process continues, CLK.sub.-- RECNSTR and its complement are generated at the Q and Q' outputs respectively. Thus, as illustrated in FIGS. 5a-5b, if the active low (CLR or ECLR) is asserted at the mid-point of the clock period and deasserted within at most half of a clock period, CLK.sub.-- RECNSTR will be symmetrical to CLK.sub.-- IN in frequency, i.e. same number of clocks, and its clock period will be 50% duty cycle symmetric, i.e. an equal amount of high and low time, thereby deskewing the entering clock.

While CLK.sub.-- RECNSTR and its complement can be generated at the Q and Q' outputs respectively with the active low asserted at the mid-point of the clock period and deasserted just before the new clock period, i.e. CLR, as illustrated in FIG. 5c, a slight drift of CLR causing the active low to be deasserted just after the new clock period will cause CLK.sub.-- RECNSTR to have at most half of the frequency of the entering clock. Thus, an active low asserted at the mid-point of the clock period and deasserted well before the end of half of a clock period, i.e. ECLR, as illustrated in FIG. 5d, is preferred. ECLR is essentially CLR with a shortened active low in each period. In fact, as will be explained in the descriptions to follow, when the proper timing for CLR is established, the proper timing of ECLR is also established.

As illustrated in FIG. 4, the two versions of active low, CLR and ECLR, are asserted and deasserted at the clear input of the first flip flop 38a by the first gate assembly 42a. The first gate assembly 42a generates CLR and ECLR based on two input signals. TAPOUT and ETAPOUT, it receives from the programmable delay chain 40. The programmable delay chain 40 generates TAPOUT and ETAPOUT based on CLK.sub.-- RECNSTR and its complement, it receives from the first flip flop 38a, i.e. the Q and Q' outputs. TAPOUT is further dependent on the delay setting selected for the programmable delay chain 40. Thus, the timing of CLR and ECLR are dependent on the timing of TAPOUT and ETAPOUT, and ultimately on the delay setting selected for the programmable delay chain 40. The delay setting for the programmable delay chain 40 is selected using the TAPSEL signal. However, because the operating characteristics of CMOS integrated circuits vary due to process, temperature, and voltage, the effective delay yield of a delay setting has to be determined operationally. Therefore, the proper delay setting for the programmable delay chain 40 is selected with a calibration process. The first gate assembly 42a, and the generation of the CLR and ECLR signals will be described in further detail later, after the programmable delay chain 40, and the generation of the TAPOUT and ETAPOUT signals have been described. The selection of the delay setting for the programmable delay chain 40 will also be described in further detail below.

The second flip flop 38b is used to generate CLK.sub.-- RECNSTR'. The second flip flop 38b is also a type D rising clock edge triggered flip flop. The second flip flop 38b generates CLK.sub.-- RECNSTR' in substantially the manner as the first flip flop 38a generates CLK.sub.-- RECNSTR, except the active low is asserted and deasserted at the clear input of the second flip flop 38b by the second gate assembly 42b. Unlike the first flip flop 38a, there is only one version of the active low, CLR', which is generated by the second gate assembly 42b based on the input signal TAPOUT it receives from the programmable delay chain 40. CLR' is generated by the second gate assembly 42b in a similar manner as CLR is generated by the first gate assembly 42a. As described earlier, the operating characteristics of CMOS integrated circuits vary due to process, temperature and voltage. In particular, the effective delay yield of a selected delay setting for the programmable delay chain 40 increases as temperature increases. As a result of the increased delay, ECLR and CLR' will be asserted after the mid-point of the clock period, thereby increasing the high time of each clock period of CLK.sub.-- RECNSTR and CLK.sub.-- RECNSTR' and causing these clocks to drift away from 50% duty cycle symmetry. Additionally, CLR' will be deasserted after a new clock period, causing CLK.sub.-- RECNSTR' to have at most half of the frequency of CLK.sub.-- IN. Thus, CLK.sub.-- RECNSTR' can be used as a basis during normal operation to provide early warning to the fact that the clock period of CLK.sub.-- RECNSTR is starting to drift away from 50% duty cycle symmetry. The second gate assembly 42b, and the generation of CLR', will also be described in further detail, after the programmable delay chain 40, and the generation of the TAPOUT and ETAPOUT signals have been described.

Referring now to FIGS. 6a-6e, six block diagrams illustrating one embodiment of the programmable delay chain of FIG. 4 are shown. As described earlier, the programmable delay chain 40 is used to generate the TAPOUT and ETAPOUT signals. As shown in FIGS. 6a-6c, the programmable delay chain 40 comprises 4 inverters 51a-51d and 40 delay elements 52a-52nn, organized into 8 delay chains. Additionally, as shown in FIG. 6d, the programmable delay chain 40 further comprises 6 multiplexors 54a-54f, coupled to the delay chains, an OR gate 50, and 4 AND gates 56a-56d coupled to the OR gate 50 and the first four multiplexors 54a-54d.

The delay elements 52a-52nn are used to apply various amounts of delays to the Q and Q' output of the first flip flop. In this embodiment, 42 delayed outputs, TAP[6:0], TAP[16:10], TAP[26:20], TAP[36:30], TAP[46:40], and TAP[56:50]are provided. Each delay element 52* is designed to provide a worst case delay resolution of about 1 to 1.5 ns. However, as described earlier, since the operating characteristics of integrated circuits vary due to process, temperature, and voltage, the delay yield of a delay element during operation is not known precisely. The number of delay elements employed and the number of delay chains organized in the programmable delay chain 40 are dependent on the delay resolution requirement of a particular high speed digital circuit. In general, the higher the maximum operating speed, the finer the delay resolution, and therefore, the larger the number of delay elements will be required.

A particular implementation of a delay element 52* comprising a number of inverters 58a-58d is illustrated in FIG. 6e. In this implementation, output OUT and L1 are simply delayed IN, whereas output L2 is delayed and inverted IN of the delay element 52*. Therefore, notwithstanding both Q and Q' outputs of the first flip flop are employed to generate TAP[6:0] through TAP[56:50], since all delays to the Q output of the first flip flop are implemented with even number of inverters, and all delays to the Q' output of the first flip flop are implemented with odd number of inverters, effectively TAP[6:0] through TAP[56:50] are all delays to the Q output of the first flip flop. While the present invention is being described with a programmable delay chain having delay elements comprising inverters, it will be appreciated that the present invention may be practiced with delay elements constituted in other well known manners.

The multiplexors 54a-54f are used to select one of TAP[56:0] for output as TAPOUT based on the delay setting TAPSEL of the programmable delay chain 40. Since the precise delay yield of each of the delay elements 52a-52nn is not known, the delay setting TAPSEL is determined through a calibration process, which will be described in further detail below. As shown in FIG. 6a, one of the earlier TAP[56:0], TAP2 in this embodiment, is preselected for output as ETAPOUT. Thus, through the preselection of one of TAP[56:0] for output as ETAPOUT, and the dynamic selection of one of TAP[56:0] for output as TAPOUT, the timing of CLR, ECLR, and CLR' are controlled.

The OR gate 50 and the AND gates 56a-56d are used to shorten the length of the active time of TAPOUT in each clock period with TAP2 (ECLR.sub.-- N=0). This is done in concert with the first gate assembly when it uses ETAPOUT (ECLR.sub.-- N=0) to shorten the length of the active time of CLR in each clock period to provide ECLR instead of CLR to the first flip flop, thereby preventing any additional delays that might be incurred by TAPOUT due to the multiplexors 54a-54f from undermining the shortening of the length of the active low of CLR in each clock period by the first gate assembly.

Referring now to FIGS. 7a-7b, two block diagram illustrating one embodiment of the first and second gate assemblies of FIG. 4 in further detail is shown. As described earlier, the first and second gate assemblies 42a-42b are used to provide CLR/ECLR and CLR' to clear the first and second flip flops respectively. Both the first and second gate assemblies 42a-42b comprise a OR gate 60a, 60b and a NAND gate 62a, 62b, coupled to each other. The OR gate 60a of the first gate assembly 42a receives ETAPOUT and ECLR.sub.-- N as inputs, whereas inputs to the OR gate 60b of the second gate assembly 42b are tied to Vdd. Each of the NAND gate 62a, 62b of the two gate assemblies 42a and 42b receives the output of its OR gate 60a, 60b, and TAPOUT as inputs.

Thus, when ECLR.sub.-- N is set at logic one (during calibration), the first gate assembly 42a asserts the active low at the clear input of the first flip flop whenever TAPOUT equals logic one, and deasserts the active low at the clear input of the first flip flop whenever TAPOUT equals logic zero, independent of the value of ETAPOUT. However, when ECLR.sub.-- N is set to logic zero (during normal operation), the first gate assembly 42a asserts the active low at the clear input of the first flip flop only when both TAPOUT and ETAPOUT equal logic one, and deasserts the active low at the clear input of the first flip flop when at least one of TAPOUT and ETAPOUT does not equal logic one. On the other hand, the second gate assembly 42b asserts the active low at the clear input of the second flip flop whenever TAPOUT equals logic one, and deasserts the active low at the clear input of the flip flop whenever TAPOUT equals logic zero, independent of the values of ETAPOUT and ECLR.sub.-- N.

Referring now to FIG. 8, a block diagram illustrating one embodiment of a counter-based frequency measuring circuit provided as one of the measurement or comparison circuits of FIG. 3 is shown. The counter-based frequency measuring circuit 44a is used to measure the ring signals and CLK.sub.-- RECNSTR during calibration, and CLK.sub.-- RECNSTR and CLK.sub.-- IN during normal operation, over a selected time period, to determine their respective frequencies. As illustrated in FIG. 8, the counter-based frequency measuring circuit 44a comprises a multiplexor 66 for selecting either a ring signal, CLK.sub.-- IN or CLK.sub.-- RECNSTR for measurement, and a number of gates, counters and latches 68 for counting the selected signal over the selected time period. The gates, counters, and latches 68 may be configured in a variety of well known manners, and will not be described in further detail. Preferably, the gates, counters, and latches 68 are configured to accommodate a variety of frequencies.

Referring now to FIG. 9, a block diagram illustrating one embodiment of the ring oscillator of FIG. 3 is shown. The ring oscillator 64 is used during calibration to generate a number of ring signals. The ring oscillator 64 comprises a selector 70, a number of delay elements 72a-72p identical to those used for the programmable delay chain, organized into two rings of predetermined length and coupled to the selector 70. Since the number of delay elements 72* in each ring is predetermined, thus by measuring the frequencies of various ring signals generated by these rings, the operational delay yield of the delay elements can be estimated.

Referring back to FIG. 3, as described earlier, preferably multiple measurement or comparison circuits 44 are provided to each TAR. In the presently preferred embodiment, the measurement or comparison circuits 44 further comprise the delay line and register-based frequency and high/low time measuring circuit described in copending U.S. patent application, Ser. No. 08/040,623, entitled Method and Apparatus for Measuring The Frequency And High/Low Time Of A Digital Signal, assigned to the assignee of the present invention, which is fully incorporated by reference herein. The frequency and high/low time measuring circuit is used to measure the frequency and high/low time of a period of CLK.sub.-- RECNSTR, during calibration and normal operation. The frequency and high/low time measuring circuit described in the copending application comprises a delay line, two registers coupled to the delay line, and a set of multiplexors coupled to the two registers. The delay line is constructed with delay elements identical to those contained in the programmable delay chain. The delay line has a number delayed outputs, each delayed output has an additional equal amount of delay applied to the input signal, i.e. the second delay output has twice the amount of delay applied to the first delay output, the third delay output has three times the amount of delay applied to the first delay output, etc. etc. Each of the two registers has a number a bits corresponding to the delay outputs of the delay line, one bit per delay output.

As applied to the present invention, the delay line receives CLK.sub.-- RECNSTR as input. The delayed outputs are sampled at two different points in time, and the sampled results are stored in the two registers. Based on the amount of elapsed time between the two samples, and the differences between the two sample results, the frequency, as well as the high/low time of a period, of CLK.sub.-- RECNSTR can be computed. For more detailed description of the circuit, see the copending application.

Still referring to FIG. 3, in the presently preferred embodiment, the measurement or comparison circuits 44 further comprise the counter-based frequencies comparison circuit described in copending U.S. patent application, Ser. No. 08/040,477, entitled Method and Apparatus for Testing Frequency Symmetry Of Two Digital Signals, assigned to the assignee of the present invention, which is fully incorporated by reference herein. The frequencies comparison circuit is used to compare the frequencies of CLK.sub.-- RECNSTR and CLK.sub.-- IN, during calibration and normal operation. Additionally, the frequencies comparison circuit is used to compare the frequencies of CLK.sub.-- RECNSTR' and CLK.sub.-- IN during normal operation. The frequencies comparison circuit described in the copending application comprises two input generating circuits, two sets of counters, and two comparison circuits. The two sets of counters are correspondingly coupled to the input generating circuits and the two comparison circuits. Additionally, the first comparison circuit is also coupled to the first and second input generating circuits.

As applied to the present invention, the first and second input generating circuits receive CLK.sub.-- IN and either CLK.sub.-- RECNSTR or CLK.sub.-- RFCNSTR' as their clock inputs respectively, and in turn use them to generate enable inputs for the corresponding counters. The counters receive CLK.sub.-- IN, either CLK.sub.-- RECNSTR or CLK.sub.-- RECNSTR', and the enable signals as inputs, and count the input clocks while the enable signals are provided. The first comparison circuit monitors the first set of counters, and stops both input generating circuits from providing further enable inputs to both sets of counters, after the first set of counters reaches a predetermined level, thereby stopping both sets of counters from further counting. The second comparison circuit monitors the second set of counters, and issues a control signal indicating the input signals are symmetric to each other in frequency, if the second set of counters also stops substantially at the predetermined level, i.e. within a predetermined acceptable threshold. In the presently preferred embodiment, when measuring CLK.sub.-- RECNSTR' and CLK.sub.-- IN during normal operation, the control signal is also used to trigger an interrupt indicating the clock period of CLK.sub.-- RECNSTR is starting to drift away from 50% duty cycle symmetry if the control signal indicates the frequency of CLK.sub.-- RECNSTR' has drifted away from CLK.sub.-- IN beyond the predetermined acceptable threshold. For more detailed description of the circuit, see the copending application.

Referring now to FIG. 10, a block diagram illustrating the method of the present invention is shown. Initially, the programmable delay chain and the first gate assembly of the digital clock reconstruction circuit are put into calibration mode by setting ECLR.sub.-- N to logic one step 90. Under the presently preferred embodiment, a large initial delay setting is selected for the programmable delay chain, by selecting one of the larger delays to the Q output of the flip flop to be output as TAPOUT, step 94. Then, various calibration measurements are taken with a maximum operating frequency calibration clock provided as CLK.sub.-- IN using the various measurement and comparison circuits provided, step 96. If CLK.sub.-- RECNSTR is not symmetric to CLK.sub.-- IN in frequency, or its period is not 50% duty cycle symmetric, no branch of step 98, the delay setting for the programmable delay chain is decreased, and steps 96-98 are repeated. Since the search for proper delay setting starts with a large delay setting and decremented towards the lower delay settings, 50% duty cycle symmetry for the reconstructed period will be achieved simultaneously when frequency symmetry is achieved between CLK.sub.-- RECNSTR and CLK.sub.-- IN. Since the calibration is performed with the maximum operating frequency, once the proper delay setting for the programmable delay chain is determined, the minimum amount of high and low time required by the TAR will be provided regardless of the ultimate operating speed.

While it is preferred to calibrate the clock reconstruction circuit starting with a large initial delay setting and decrement the delay setting systematically, it will be appreciated that the clock reconstruction circuit may be calibrated with other initial delay settings and other manners of systematically varying the delay setting. In particular, the clock reconstruction circuit may be calibrated starting with a small initial delay setting and increment the delay setting systematically as long as CLK.sub.-- RECNSTR is symmetric to CLK.sub.-- IN in frequency. Under this approach, the proper delay setting is the delay setting just prior to the delay setting that causes CLK.sub.-- RECNSTR to have at most half of the frequency of CLK.sub.-- IN (CLR being deasserted just after a new clock period). At this delay setting, frequency symmetry between CLK.sub.-- RECNSTR and CLK.sub.-- IN, and 50% duty cycle symmetry for the reconstructed clock period are also simultaneously achieved.

Once frequency symmetry between CLK.sub.-- RECNSTR and CLK.sub.-- IN and 50% duty cycle symmetry for the reconstructed period are achieved, yes branch of step 98, the programmable delay chain and the first gate assembly of the digital reconstruction clock is put into operation mode by setting ECLR.sub.-- N to logic zero step 104. Various operational monitoring measurements are taken and analyzed using the various measurement and comparison circuits provided, step 106. If at any time, the clock period of CLK.sub.-- RECNSTR is starting to drift away from 50% duty cycle symmetry, yes branch of step 108, an interrupt is triggered. Otherwise, no branch of step 108, steps 106 and 108 are repeated. It will be appreciated that the entire calibration process 88 may be controlled by either hardware or software.

In the presently preferred embodiment, the selection of the initial large delay setting in step 94 is guided by the delay yield estimate for the delay elements of the programmable delay chain. The delay yield is estimated by measuring the frequencies of the ring signals generated by the ring oscillator using the counter-based frequency measuring circuit. In step 96, the frequency of CLK.sub.-- RECNSTR is measured using the counter-based frequency measurement circuit as well as being compared to CLK.sub.-- IN using the frequencies comparison circuit. The high and low time of the reconstructed period is also measured, notwithstanding 50% duty cycle symmetry in the reconstructed period will be achieved simultaneously when frequency symmetry is achieved between CLK.sub.-- RECNSTR and CLK.sub.-- IN under the calibration process of the present invention. In step 106, CLK.sub.-- RECNSTR is also measured and compared to CLK.sub.-- IN periodically using the various measurement and comparison circuits, in addition to measuring and comparing CLK.sub.-- RECNSTR' to CLK.sub.-- IN periodically using the frequencies comparison circuit.

Additionally, because CLK.sub.-- RECNSTR' is used as a basis to provide early warning to the fact that the clock period of CLK.sub.-- RECNSTR is starting to drift away from 50% duty cycle symmetry, a delay setting slightly smaller than the optimal delay setting is selected for the programmable delay chain to provide a small 50% duty cycle symmetry drifting range for the clock period of CLK.sub.-- RECNSTR without triggering the early warning.

Alternatively, steps 94-102 may be performed by applying a preset to the first flip flop of the digital clock reconstruction circuit, in lieu of providing a maximum operating frequency calibration clock as CLK.sub.-- IN. The digital clock reconstruction circuit is functionally transformed into a ring oscillator. Under this approach, steps 96-102 are repeated until the maximum operating frequency is measured for CLK.sub.-- RECNSTR. At this time, 50% duty cycle symmetry for the reconstructed period will also be achieved simultaneously. However, under this alternative, the clear input of the flip flop must have priority over its preset input.

While the present invention has been described in terms of presently preferred and alternate embodiments, those skilled in the art will recognize that the invention is not limited to the embodiments described. The method and apparatus of the present invention can be practiced with modification and alteration within the spirit and scope of the appended claims. The description is thus to be regarded as illustrative instead of limiting on the present invention.