BACKGROUND '

This disclosure relates to synchronizing circuits; and more particularly, it relates to synchronizers which sample an asynchronous digital input signal with a clock and operate with low error rates.

Synchronizers, of the type with which the present invention is concerned, have use whenever digital informa¬ tion is to be transferred between two digital modules that operate asynchronously to one another. For example, digi- tal computers usually have their own internal clock to which their internal operations are synchronized; and thus, when one computer sends information to a second computer, the information will arrive at the second computer asynchro¬ nously with respect to its internal clock. Similarly, when information is entered by an operator via a keyboard into a computer, the keys are pressed asynchronously with respect to the computer's internal clock.

In the above cases, a synchronizer circuit is often used to sample and hold the incoming data until it can be received by the receiving computer. However, since the incoming data is asynchronous, there is a certain proba- bility that the data will change at the same time that it is being sampled. And, when that happens, the sample that is taken may be at an intermediate voltage level which lies between a full "1" and a full "0".

Then, if the intermediate voltage is operated on by the receiving computer, an error can occur. This is because the intermediate voltage can be interpreted by the computer as a "1" when it actually was a "0", and vice versa. Consequently, it is important that the synchronizer circuit be designed to reduce the probability with which such intermediate voltage errors occur.

One synchronizer circuit of the prior art which is related to the present invention is described in U.S. Patent 3,953,744 which issued in 1976 to Kawagoe. However, with the '744 synchronizer, the probability of an inter- mediate voltage error occurring is too high; and this will be explained in detail in the Detailed Description in conjunction with FIG. 10.

Accordingly, it is a primary object of the invention to provide an improved synchronizing circuit which greatly reduces intermediate voltage errors.

INTENTIONALLY LEFT BLANK

INTENTIONALLY LEFT BLANK

BRIEF DESCRIPTION OF THE DRAWINGS Various features and advantages of the inventio are described herein in conjunction with the accompanyin drawings wherein:

FIG. 1 illustrates a preferred embodiment of th invention; FIG. 2 illustrates the overall operation of th

FIG. 1 embodiment;

FIG. 3 is a detailed circuit diagram of on version of the FIG. 1 embodiment;

FIG. 4 is a modification of the FIG. 3 circui that is used to analyze one of the feedback loops in th FIG. 3 circuit;

FIG. 5 is a set of equations which describe th transient operation of the FIG. 4 circuit;

FIG. 6 is a modification of the FIG. 3 circuit which is used to analyze a second feedback loop in the FIG. 3 circuit;

FIG. 7 is a set of equations which describes the transient operation of the FIG. 6 circuit;

FIG. 8 illustrates the transient response of both feedback loops in the FIG. 3 circuit;

FIG. 9 is a set of equations which calculate the MTBF of the FIG. 3 circuit; and FIG. 10 is a set of equations which calculate the

MTBF of a prior art circuit.

DETAILED DESCRIPTION

Referring now to FIG. 1, the details of a preferred embodiment of the invention will be described. This embodiment includes a Schmidt trigger 10, an inverter 11, and a resistor 12. All of these components are inter¬ connected in a loop as illustrated. Also included is a transistor 13 which is coupled to the input terminal of the Schmidt trigger 10, and a transistor 14 which is coupled to the output terminal of the Schmidt trigger 10.

In operation, a digital input voltage v^ which is to be synchronized is applied to the source of transistor 13, and a digital clock CK^ is applied to the gate o ^~ transistor 13. When clock CK_ is high, a sample of the voltage v]_ is transferred to the input terminal of the Schmidt trigger 10. This sample is indicated as voltage V2« Then, components 10, 11, and 12 operate on the voltage V2 to produce a voltage V3 on the output terminal of the Schmidt trigger. Thereafter, a digital clock CK2 is applied to the gate of transistor 14 to transfer the voltage V3 to the drain of transistor 14 as a voltage V4.

Now in the above-described operation, the voltag V _] _ can change at any time with respect to the clock signa CK _] _. This is illustrated in FIG. 2. There, voltage v changes before the first clock pulse 20 and remains stabl during the clock pulse; whereas during the second cloc pulse 21, the voltage v ^* j_ changes state. When voltage stays stable during a CK^ clock pulse, the other voltage V2 and V3 stabilize shortly after the rising edge of th CK _j clock. This is indicated by the transitions 20a an 20b in FIG. 2. But when voltage vi changes state during CKi clock pulse, then the voltages V2 and V3 take a longe time to stabilize, as is indicated by the transitions 21 and 21b in FIG. 2.

In the worst case, the voltage _ changes jus before the clock CK]_ goes low. Suppose, for example, tha a full "1" voltage is five volts and a full "0" voltage i zero volts; and suppose further that voltage ^ change just before clock CK^ goes low. In that case, the voltag sample V2 can have any magnitude between zero and fiv volts. This intermediate voltage is then operated on b components 10, 11 and 12 to produce voltage V3 as a ful "1" voltage and regenerate the voltage V2 as a full "0 voltage; or vice versa.

To further explain the above regenerative process the detailed transistor makeup of one preferred version o the components 10 and 11 will now be described in conjunc tion with FIG. 3. There, the Schmidt trigger 10 consist of one P channel transistor 10a and three N channe transistors 10b, 10c, and lOd. Also the inverter 1 consists of one P channel transistor 11a and one N channe transistor lib. All of these transistors, together wit resistor 12, are interconnected as illustrated.

Inspection of the FIG. 3 circuit shows that it contains two separate feedback loops. One feedback loop includes resistor 12 together with transistors 11a and lib; and they are coupled as shown from the output terminal of the Schmidt trigger 10 to the input terminal of the Schmidt trigger. In operation, this feedback functions to slowly regenerate the voltage 2 on the input terminal of the Schmidt trigger 10 in inverse proportion to the voltage V3. This prevents the V2 voltage sample which transistor 13 provides from dissipating due to leakage.

By comparison, the other feedback loop includes transistors 10c and lOd; and they are coupled as shown from the output terminal of the Schmidt trigger 10 to the drain of transistor 10b. In operation, this feedback functions to quickly increase the voltage V3 when the voltage sample V2 is below a certain predetermined level, and vice versa, without altering the magnitude of the voltage sample V2 on the input terminal.

For example, suppose that the voltage sample V2 is at an intermediate low voltage which is near the midpoint between a full "0" voltage and the full "1" voltage. In that case, transistors 10a, 10b, and lOd will all be conducting about the same current; and transistor 10c will just be starting to turn on. As transistor 10c starts to conduct, the voltage across it will drop; and consequently, the source to drain voltage across transistor lOd will increase. That in turn will lower the gate to source voltage of transistor 10b, and thus transistor 10b will conduct less. Lowering the conductance of transistor 10b will raise the voltage V3, which in turn will cause transistor 10c to conduct more; etc. Thus, by repeating this feedback action, transistor 10b will turn off and voltage V3 will reach a full "1" level.

An analytical expression for the speed with whic the above feedback action occurs will now be derived i conjunction with FIGs. 4 and 5. FIG. 4 is a modifie version of FIG. 3 in which the transistors 11a and lib ar deleted, and certain parasitic capacitors C3, C3a, and C are added. Deletion of the transistors 11a and lib enables the speed of the transistor 10c feedback to be analyze separately from the other feedback. Also, the parasitic capacitors are added since they are inherently present at the input of a transistor's gate and the output of transistor's source and drain.

Inspection of FIG. 4 shows that a current i3 through transistor 10a will charge the parasitic capacitor C3 and thereby raise voltage V3. Similarly, a current i3 _a through transistor 10c will charge the parasitic capacitor C3a and thereby raise the voltage V3 _a across transistor lOd. This charging action is stated mathematically in FIG. 5 by the differential equations 1 and 2. There, symbol g^ represents the transconductance of an N channel transistor when its gate voltage is midway between a full "1" voltage and a full "0" voltage.

Equations 1 and 2 may be solved by various tech¬ niques, such as by Laplace transforms; and their solution yields equation 3. It states that when the voltage V2 is near the midpoint between a full "1" voltage and a full "0" voltage, the voltage V3 will vary due to the transistor 10c feedback as an exponential. This exponential has a time constant , where ^~ ~~- is a function of the parasitic capacitors C3 and C3a, and the transconductance g^. Capacitor C3 is the capacitor which current i3 must charge in order to raise the voltage V3. Inspection of FIG. 4 shows that capacitor C3 is equal to the output junction capacitance of two transistors (transistors 10a

an 10b) and the gate capacitance of one transistor (transistor 10c). This is stated by equation 4. Typi¬ cally, the gate capacitance of a transitor is about three times the transistor's output junction capacitance, and thus equation 4 can be rewritten as equation 5. Similarly, inspection of FIG. 4 shows that the capacitor C3a is equal to the output junction capacitance of three transistors (transistors 10b, 10c and lOd). This is stated by equation 6. Substitution of equations 5 and 6 into equation 3 yields the equation 7 expression for the time constant Ti¬ lt should be pointed out that in the above analysis, the parasitic capacitor C (which equals the gate capacitance of the transistors 11a and lib) is ignored. That, however, does not make the analysis inaccurate because capacitor C together with resistor 12 form a low pass filter. Consequently, the voltage V3 can quickly vary without having to change the voltage across capacitor C ^r .

Turning now to FIGs. 6 and 7, the speed by which the second feedback loop operates in the FIG. 3 circuit will be analyzed. FIG. 6 is a modification of the FIG. 3 circuit in which the feedback through transistor 10c is disconnected. This allows the operation of the feedback through transistors 11a and lib to be analyzed independent of- the feedback which transistor 10c provides. Also in FIG. 6, resistor 12 is short-circuited. This is done to simplify the analysis. Shorting resistor 12 will produce a faster response time; however, even this faster response will be slower than the response of the FIG. 4 loop. Further in FIG. 6, parasitic capacitors C2 and C* are added.

Inspection of FIG. 6 shows that a current i3' through transistor 10a charges the parasitic capacitor C in order to raise the voltage V3. Similarly, a current i2

through transistor lib discharges capacitor C2 in order t lower the voltage V2» This is stated mathematically i FIG. 7 by equations 10 and 11. There, g^ is the transcon ductance of an N channel transistor and gp is the transcon ductance of a P channel transistor when its gate voltage is midway between a full "1" voltage and a full "0" voltage.

Solving equations 10 and 11 yields equation 12. It states that when a voltage sample V2, which is near the midpoint between a full "1" voltage and a full "0" voltage, is placed on the capacitor C2, the output voltage V3 will change as an exponential due to the feedback that occurs through the transistors 11a and lib. And, that exponential has a time constant ' ^" 2 which is a function of C2, C, g^ and gp• Time constant ~~ can be simplified by utilizing equations 13, 14 and 15. Equation 13 states that the transconductance of an N channel transistor is approxi mately twice the transconductance of a P channel tran¬ sistor. Equation 14 states that the parasitic capacitor C' is equal to the gate capacitance of three transistors (transistors 10c, 11a, and lib) plus the output junction capacitance of two transistors (transistors 10a and 10b). Similarly, equation 15 states that the parasitic capacitor C2 is equal to the gate capacitance of three transistors (transistors 10a, 10b, and lOd) and the output junction capacitance of two transistors (transistors 11a and lib).

Substitution of equations 13, 14, and 15 into equation 12 yields equation 16. It gives an expression for the time constant 1~ ^~ 2 - ⁿ terms of the junction capacitance and N channel transconductance. Comparison of equation 16 with equation 7 yields equation 17, which shows that the time constant " "2 - ^s much larger than the time constant ~~ .

This result is illustrated graphically in FIG. 8. There, a curve 31 shows the transient response of the feedback loop which passes through transistor 10c; and a curve 32 shows the transient response of the feedback loop which passes through transistors 11a and lib.

Of primary importance in the disclosed synchro¬ nizer circuit is the speed with which voltage V3 changes. In particular, in order to reduce logic errors in external circuitry which uses the synchronizer, it is critical that the voltage V3 does not stay very long at any intermediate voltage which lies between the "0" and "1" voltage levels. By comparison, the time that is spent by the voltage V2 between the "0" voltage level and the "1" voltage level is relatively unimportant since that voltage does not go to any external circuitry.

An equation which mathematically relates the time constant 7~s ^{of an} Y synchronizer to its MTBF (mean time between failure) is given as equation 20 in FIG. 9. This equation is derived at pages 361-363 of the book entitled The Design and Analysis of VLSI Circuits by Glasser et al which is published by Addison-Wesley, 1985. In equation 20, fcKl i ^{s the} frequency of clock CK ; f^ _n is the frequency of transitions in the signal v_; Δ_is the time which it takes the input signal to make a transition through the intermediate voltages that lie between a "0" and a "1"; T is the time delay that occurs between the placing of a voltage sample on the input terminal of the synchronizer and the use by external circuitry of the synchronizer output voltage; and 7s i- ³ t e time constant of the synchronizer.

Suppose, for example, that fcκi ^{= 10} MHz, f _in = 1

MHz, Δ= 1 nanosecond, and T = 30 nanoseconds. This i stated by equation 21. Then, substituting equation 21 int equation 20 yields equation 22. There, the only paramete that remains is the synchronizer time constant TJs .

In the disclosed synchronizer, the two feedbac loops of FIGs. 4 and 6 operate in parallel. Consequently the time constant of the disclosed synchronizer as a whol will be less than each of the individual time constants "J and ~2• This is stated , by equation 23. However, t simplify the analysis, assume that the combined tim constant merely equals ~ ι - Then, substituting ~ ι int equation 22 yields equation 24.

Assume now some reasonable values for the parame ters g^ and Cj in equation 24. Such values are given b equation 25. They are representative of a transisto having a one micron gate length and a sixteen micron gat width. Substituting equation 25 into equation 24 yield equation 26, which states that the MTBF for the disclose synchronizer is over 20,000 years!

By comparison, consider now the MTBF which wil result from the Glasser et al synchronizer which consist of two inverting logic gates that have their respectiv output terminals and input terminals connected together i a loop. Such a synchronizer can be analyzed by referrin back to FIG. 6 and there eliminating transistors 10c an lOd, and grounding the drain of transistor 10b.

With those modifications, all of the analysis o equations 10, 11, 12, and 13 still applies. But, th capacitor C would now be expressed as equation 30 in FIG. 10; and the capacitor C2 would be expressed as equation 31 In equation 30, the two gate capacitances are due t transistors 11a and lib, and the two junction capacitor

are due to transistors 10a and 10b. Substituting equations 30 and 31, as well as equation 13, into equation 12, yields equation 32.

Next, equation 32 can be substituted into equation 22 to thereby obtain an expression for the MTBF of the synchronizer which consists of two inverters. This result is given by equation 33. Then, the previous values of g _N and Cj as given by equation 25 can be substituted into equation 33. This yields equation 34, which states that the MTBF of the two-inverter synchronizer is only about 0.5 yearl

As a further comparison, consider the synchronizer which is described in U.S. Patent, 3,953,744. It includes two inverting logic gates that have their respective output terminals and input terminals connected together in a loop (just like the Glasser et al synchronizer); and in addi¬ tion, a Schmidt trigger is included which passes the input signal that is to be synchronized to the two inverting logic gates. This is shown in FIG. 1 of patent '744. Now in the '744 synchronizer, the only effect of the Schmidt trigger is to shorten the rise and fall times of the signals that are sent to the cross-coupled logic gates. In other words, the Schmidt trigger merely decreases the term Δthat is shown in FIG. 8 and is part of equation 20 in FIG. 9. The "744 Schmidt trigger does not, and cannot, affect the time constant of the cross-coupled logic gates since it is outside of the loop which they form.

In the above analysis of the disclosed synchro- nizer, and the Glasser et al synchronizer, Δ was assumed to equal one nanosecond (see equation 21) . Feeding a signal with a one nanosecond rise and fall time into a Schmidt trigger would produce an output signal whose rise and fall

time is decreased at the very most by a factor of about te to 0.1 nanoseconds. Thus, the MTBF of the '744 synchro nizer would merely be increased by the same factor of ten to 5.0 years. A preferred embodiment of the invention has no been described in detail. In addition, however, man changes and modifications can be made to these detail without departing from the nature and spirit of th invention. For example, in the embodiment of FIG. 3, th Schmidt trigger 10 and inverter 11 are made of CMO transistors. But as an alternative, the Schmidt trigger 1 and inverter 11 can also be made of bipolar transistors, o entirely of N channel transistors.

Also in the FIG. 3 embodiment, the particula circuit arrangement for the two feedback loops can be modi fied. In general, all that is required is that the firs feedback loop operates to quickly increase the outpu voltage 3 when the voltage sample V2 is below a pre determined level, and vice versa, without altering th voltage sample V2; and that the second feedback loo operates slowly to modify the voltage sample V2 in invers proportion to the output terminal voltage. Preferably, both of the feedback loops respond exponentially to voltage sample on the input terminal with the time constant of the first feedback loop response being t least twenty percent smaller than the time constant of the second feed back loop. Also preferably, this difference in the time constants is achieved by making the capacitance of the first feedback loop substantially smaller than the capaci tance of the second feedback loop.

Accordingly, it is to be understood that the invention is not limited to the above details but is defined by the appended claims.