LIBRARY OF THE UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN 510.84 ho.£|-8>0 *3 CO Digitized by the Internet Archive in 2013 http://archive.org/details/theoryofasynchro75mull UNIVERSITY OF ILLINOIS GRADUATE COLLEGE DIGITAL COMPUTER LABORATORY REPORT NO. 75 A THEORY OF ASYNCHRONOUS CIRCUITS I D. E. Muller and W. Scott Bartky November 7 > 1956 This work has been supported in part by the Office of Naval Research under Contract NR 0^ 001 1. Introduction 1.1 Switching circuits are commonly classified as being either synchro- nous or asynchronous, depending upon whether or not the signals in the circuit are synchronized with some source of fundamental frequency (or clock) which regulates the entire circuit. Much of the mathematical theory of switching circuits and many design techniques have been limited to synchronous circuits, because it is possible to predict the state (or condition) of a synchronous circuit at the time of any given clock signal if one knows the initial state of the circuit and its logical characteristics. As a result, many design problems may be handled by logical algebra. However, in an asynchronous circuit it is not possible to predict the state immediately resulting from a previous state from a knowledge of the logical characteristics of the circuit alone. The resulting state may also depend upon the relative speeds of some of the logical elements which comprise the circuit. The possibility of such "races" in asynchronous circuits complicates the theory unless one is prepared to make definite assumptions concerning element speeds, and thus reduce the problem to the synchronous case, or else design one's circuits so that only one logical element is able to change at a time. 1.2 We shall not attempt here to extend the theory of synchronous circuits to cover the asynchronous case, but rather develope a comprehensive theory of asynchronous circuits which will differ from the synchronous theory in many fundamental respects. One of our main objects will be to describe the properties of circuits which are speed independent in the sense that their ultimate behaviors do not depend upon the relative speeds of their elements. What is meant by "ultimate behavior" will be set down precisely in the later discussion and techniques for designing speed independent circuits will be described toward the end. 1.3 Existing theories of asynchronous circuits have been mostly concerned with what might be called the "black box" model of a circuit. It deals with a "black box" or circuit having a number of inputs, a number of outputs, and a number of internal states. When the input signals are changed, the circuit proceeds from its existing internal state to another uniquely determined internal state and the outputs change accordingly. One must always assume that no further change is made in the inputs until this second internal -1- If ll Hi ii ii state has actually "been achieved, for otherwise the behavior of the circuit would be indeterminate. The theory treats the way the states and outputs change as a result of manipulating the inputs and in most respects is similar to the corresponding theory of synchronous circuits . 1.4 The theory presented here is not a revision of the "black box" theories but may be regarded as a supplement to them since it is concerned with the changes which the circuit undergoes if the inputs, if any, are held fixed. It therefore deals with those transitions which occur between input changes. It might also be used to describe the action of a circuit which has no inputs of the type referred to above; for example, an automatic computer. Since the inputs, if they exist, are regarded as fixed in the present theory they need not appear explicitly in our description of a circuit and may be regarded as parameters affecting the circuit's logical properties. 2 . Description of a Circuit 2.1 In an idealized switching circuit the physical signal present at any point within the circuit may assume one of a discrete set of values. This physical signal may take the form of a voltage, current, position, or other measurable quantity. Let us assume that n such measurements are sufficient to specify the state of the circuit so that any other measurements may be uniquely determined from the given n, These n signals will be denoted by symbols such as z. ,z , ... ,z and the n-tuple z = (z , z , ... , z ) will be taken as representing the state of the circuit. 2.2 Each signal z. may assume one of a discrete set of values S., where the elements of S. are finite in number and totally ordered. They may therefore be set into correspondence with the integers 0,1, ... ,k. -1, where it is assumed that S contains k elements. The set S of states is, thus, also finite and contains all n-tuples z whose components z. are taken from the sets S. . l 2.3 An important special case is that of the binary switching circuit in which a signal may take one of but two possible values so that k.=2 for each i=l,2, ... ,n. Thus, in the binary case the signal z. may be either or 1, while in the general case, it may be any member of the ordered set of values -2- 0,1, ... ,k -1. The binary case is so important that it would hardly be worth- while treating the more general case were it not possible to do so with very little additional difficulty. 2 .k In a relay switching circuit the signal z. may be conveniently taken as describing the position of the i'th relay. If this relay is of the usual two position (open or closed) type, the signal will be binary while a relay having k. standard positions could be described by the more general type of s ignal . 2.5 In an electronic switching circuit the signals are usually interpreted as voltages, and although it is true that voltages in a circuit change in a continuous fashion, one may assume that each signal value merely specifies a range of voltage rather than an exact voltage Naturally, we may only use this simplified treatment of the signals if the behavior of the circuit is adaquately described by the discrete signals. 2.6 The switching properties of an idealized switching circuit can be expressed as a set of functions ; ' = f . (z. ,z_, ... ,z ) ; 1=1,2, 1 1 1 2 n 3 ' ' where the dependent variables z.' lie in the sets S.. Each z.' is assumed to be 1 11 uniquely determined by the state z=(z, ,z , . . . ,z ). In a binary circuit these functions will be Boolean functions since all signals are assumed to be either or 1. In an electronic circuit these functions represent individual logical elements which form the circuit and serve to describe their interconnections, while in a relay circuit these functions express the behavior of the networks of contacts . 2.7 Our theory will therefore be concerned with the following concepts: (2:1a) n signals z. 5 1=1,2, ... ,n, each taken from a set S. of k. ordered elements, which will be written 0,1, ... ,k.-l. (2:1b) The set S of n-tuples of signals z=(z, ,z p , ... ,z ), called states . (2:1c) n functions z.' = f . (z ,z p , ... ,z ) which define a mapping of S onto itself o A more general switching circuit will also be influenced by outside parameters or "inputs" which should properly be included in equations (2:1c). Since the present ■3- theory is only concerned with the behavior of circuits during the time that such inputs (if present at all) are held fixed, we need not include them explicitly in the functions (2:1c). 3° Circuit Behavior 3.1 We now wish to set down a mathematical description of the behavior of an asynchronous, or non-clocked, switching circuit* If the circuit is placed in a state a=(a.,,a , ... ,a ), some of the signals will generally tend to change and the circuit will pass to another state b=(b,,b , . . . ,b ) . If the circuit is synchronous this "next" state b will occur just one clock interval later and will be uniquely determined by the functions (2:1c). In fact, b will be just a'= (a ' ,a ' , ... ,a '). In an asynchronous circuit, on the other hand, there is no clock, or source of fundamental frequency, and the relative times which the various signals take to change will depend on physical properties of the circuit elements and will be unrelated to the logical properties expressed in (2 sic). The state b is, thus, not always uniquely determined by the functions (2?lc) in the asynchronous case and, in general, may be any one of several states depending on these relative speeds. 3°2 We wish to make no specific assumptions concerning relative speeds of circuit elements in the asynchronous case and we shall try to describe all possible states b into which the circuit may pass from a given state a. This concept is expressed in terms of a relationship 0c . (3sl) a^-b means that each signal b. of b satisfies either a.< b.< a „ ' or a.> b.> a . ' , depending upon whether a.< a. ' or a.> a. ' . i-* 1— 1 1— 1— 1 * 1—1 1—1 Clearly a$,a and a&a' , but, in general, there may be other states b as well which satisfy a Jib, and in asynchronous circuit theory all such "next" states will be regarded as possible. If one thinks of a. as the existing signal and a.' as a new signal toward which point i is tending, then (3^1) may be thought of intuitively as saying, that in the "next" state b the signal b, at point i may have changed so that it lies anywhere in the closed interval a. to a.' (3:2) A partial allowed sequence of k states a(l),a(2), ... ,a(k) is any sequence of k > 1 states which, whenever k > 2, satisfies a(j)02. a(j+l) and a(j) + a(j+l) for all j=l,2, ... ,k~l. 3«3 A partial allowed sequence is a sequence of states which is intended to describe the possible behavior of a circuit which starts in state a(l) and passes through succeeding states a (2), ... ,a(k). The circuit may go on to other states after reaching a(k) and so to complete the description of the circuit's behavior we require? (3*3) An allowed sequence of states a(l),a(2), ... (which may be either finite or infinite) is any sequence satisfying: (3^3a) a ( j ) (X. a(j+l) and a(j) \ a(j+l) for all consecutive pairs of states a(j),a(j+l) in the sequence. and (3 Ob) No signal a.(j) may satisfy either a . ( j ) =a.(k) j or a. (j ) = a. (k) > a' . (k) for ail signals a. (k) with k > j. 3 A An allowed sequence may be regarded as a possible sequence of states through which an asynchronous circuit might pass if placed initially in state a(l) This definition (3° 3) was chosen because it was thought to best represent asynchronous circuit behavior. Other possible definitions may be more convenient from a mathematical standpoint but would fail to yield a realistic theory „ Condition (3»3b) requires that no point i in the circuit may retain a signal a . ( j ) indefinitely if it is tending toward some other value a.'(k) which remains either greater than a. ( j ) or less than a. (j) for the rest of the sequence » We note that (3*3b) permits a. (j) to remain fixed indefinitely if a'.(j) is alternately greater and less than a.(j)o (3*^0 Any finite, consecutive, set of states in an allowed sequence forms a partial allowed sequence. (3»5) If both a(l),a(2), ... ,a(k) and a(k),a(k+l), ... ,a(m) are partial allowed sequences, then a(l),a(2), » . . ,a(k), ... ,a (m) is a partial allowed sequence. (3^5b) If the sequence a(k),a (k+l), » . . is an allowed sequence and a(l),a(2), o.. ,a(k) is a partial allowed sequence, then the composite sequence a(l),a(2), ... ,a(k), ... is an allowed sequence. (3s 6) Allowed squences may be either finite or infinite, but if a finite sequence a(l),a(2), . = . ,a(m) is allowed, then the terminal state a(m) must be an equilibrium state, i.e. one for which a(m) = a'(m) -5- l! " H H ii i i ii ■ i II 1 1 iii t i ' I ' I I' 1 i' I' \\ I The statements above follow from the definitions of allowed and of partial allowed sequences. (3=6) must hold in order to avoid a violation of (3s3b)» (3* 3b) also restricts infinite allowed sequences, but in a more subtle way as will be demonstrated later. 3-5 If an allowed sequence does indeed represent a possible sequence of states through which the circuit might pass, then it should be possible to show that at least one allowed sequence exists which starts with any state a(l) as the initial state . If no allowed sequence exists then we must seek someother way of describing the behavior of the circuit when placed in state a(l). (3: A) Given any state a(l), there must exist at least one allowed sequence a(l),a(2), . .. having a(l) as its initial state. Proof ; Form the sequence a(l),a(2), ... by the rules (3:7a) Let a(j+l) = a'(j) if a'(j) + a(j). (3? 7b) Let the sequence terminate with a(j) if a T (j) = a(j). That this sequence satisfies (3:3a), may be seen from the property a(j) 83. a' ( j ) . (3? 3b) may also be seen to hold if one notes that either a. (j) = a ' ( j ) or else a. (j) 4 a « ' («J) = a -t (j + l) so "that either a. (k) = a ' (k); with k = j, or else a ± U) 4 a i (k); with k=j+l. 3.6 In the relationship a (R. b we have expressed the notion that state b may come after state a with no intervening states , We now go on to define a new relationship a # b which will indicate that b may come after a but that intervening states may separate them. (3?8) a 5b, a is followed by b, means that a partial allowed sequence exists having a as its first and b as its last state . Two properties of the a relationship may be immediately noted . (3?9a) a &a for all states a . We note that any state a by itself forms a partial allowed sequence. (3s 9b) a <£ b and b , we have a # b. (3°D) The ^ relationship defines a partial ordering between equivalence sets . Proof : We must show? (3:lta) A & A for all A. (y.lkb) A# Band B# C implies A# C O^c) A ^ B and B 5- A implies A = B. These three properties may be easily deduced from (3:13) and (3°C). 3«8 Since the number of equivalence sets is finite and partially ordered, there must exist at least one maximum or final set. (3:15) M is a final set if and only if there is no set M* different from M, such that M # M*. (3°l6) Given any equivalence set A there must exist at least one final set M such that A <£ M. If we deny (3:l6) we may construct a sequence of sets A, A(l), A (2), ... of indefinite length such that A(j) & A(j+l) and A(j) 4 Mj+l), and involving no repetitions because of partial ordering. This requires an infinite number of sets. Similarly, a minimum set may be shown to exist, but such sets will have little importance in the present theory. -7- (3?E) a = a' (i.e. a is an equilibrium state) if and only if its equivalence set A is a final set containing just one state. Proof : If a = a', then a 5£b implies a = b. Hence A is a set containing just one state, and consequently is a final set since A ^ B would imply a ^ b for a state b in B. If A is a final set containing just one state a then since a #? a' we have either a' in B with A «£ B which is impossible or else a' in A which means a = a' . (3j17) A pseudo-final set P is an equivalence set of states p(l), ••• >p(r) which is not final and for which there is no signal index i such that p.(j) is constant over all j = 1, ... ,r and either j, P i (j ) < P i ' (j ) for all j = lj ... ,r or P 1 (J) > Pi'Cj) for all j =1, ... ,r We note that a pseudo-final set must contain more than one state, (3?F) Any allowed sequence a(l),a(2), ... must contain a state a(m) such that all states a(m),a(m+l), ••• lie in the same equivalence set T (called the terminal set of the sequence), which is either final or pseudo-final. Proof ; Since the number of equivalence sets is finite, we see that the number of sets which are represented in any allowed sequence is also finite. Hence there must be a state a(m) such that all equivalence sets in the entire allowed sequence are represented in the partial allowed sequence a(l),a(2), «.. ,a(m). Now if a(m) is in T, then all states following a(m) must also be in T. Otherwise, partial ordering of the sets would be violated. We must now show that T is either final or pseudo-final. Assume T is neither. In this case we have either a 1 (m) = a.(k) < a.'(k) for some i and all k > m or a.(m) = a.(k) >a.'(k) for some i and all k > m and consequently we have violated (3? 3b). 3-9 In view of (3s l6) we note that if any initial state u is chosen, we may always form an allowed sequence starting with u whose terminal set is final. A -8- somewhat stronger result may be proved, however. (3:G) If u is any state and U is its equivalence set, then if T is any final or pseudo-final set such that U & T, an allowed sequence may be constructed starting with u whose terminal set is To Proof : If T is a final set and t(0) is any state in T, we may form a partial allowed sequence u, ... , t(o) by (3^8) and (3^C). An allowed sequence starting with t may now be formed by the method of (3'A). Since T is final, the sequence must be entirely in T for if some member of the sequence lies in another set T, we would have T c? T*. Now if we combine these two sequences by (3°5b), w e obtain the desired sequence u, ... ,t(0), ... If T is a pseudo-final set, the same proof may be used provided we may form an allowed sequence starting with t(o) and lying entirely in T. This is done in the following paragraph . Let t(0),t(l), . .. ,t(k) be the states in the pseudo-final set T. Construct k + 1 partial allowed sequences t(0), ... ,t(l) ; t(l), . .. ,t(2) ; t(2), ... ,t(3) % ?.. J t(k-l), ... ,t(k) 5 t(k), ... ,t(0) ; by use of (3:8). Since the states at the ends of these partial allowed sequences match they may be combined by (3°5a) into a single partial allowed sequence t(0), «o. ,t(l), ooo ,t(2), . .. , ... ,t(k), ... ,t(0) contaxuing all states in T> Let us now repeat this partial allowed sequence indefinitely always letting the last member of the previous partial allowed sequence be the first member of the next. This resulting sequence is allowed since all states in T are present and (3:1?) gives us (3s 3b), k. An Examp le of a Circuit k ol As an example, consider the binary circuit whose defining equations (2:1c) are the Boolean equations (k'.l) z ± ' = z 2 v z^ V =I i z 3* = \ ( z i v z 2^ z k = S This circuit, if represented by a logical diagram, would consist of two "not" elements, an "or" element, and a special element for producing the function -9- V ll i ! r I z, (z. V z ) They are to be connected as in figure 1, where 4 1 , ^ W T Figure 1 E represents the special element. 4.2 If one writes the W relationships between the sixteen states and the corresponding J% relationships it can be shown that there are four equivalence sets. They may be designated by A,B,C, and D as follows. A contains (0000), (0010), (0011), (0100), (0110), (Olll), (1000), (1011), (1100), (1111) B contains (0001), (0101) , (1001), (1101) G contains (1110) D contains (1010) We have A^B, A^C, A^D, and C ^D„ We note also that B and D are final sets, while A is pseudo-final and C is neither final nor pseudo-final. Since (1010) is an equilibrium state, we see by (3 : ^) that its set contains no other state and must be final, 4.3 If we start with the initial state (llll) we may illustrate some of the possible allowed sequences. (4:2) (4:3) (4:4) (4:5) (1111), (1010) (1111), (1110), (1010) (1111), (1101), (1001), (0001), (0101), (1101), (1001), ... c (1111), (0111), (0110), (0100), (1100), (1000), (0000), (0010), (0110), (0100), ... . The last two allowed sequences are infinite having cycles of four and six states respectively. Sequences (4:2) and (4:3) have the final set D as their terminal set. Since this set contains just the equilibrium state (1010) we see that both these allowed sequences are finite. The sequence (4:2) proceeds directly from A to D while (4:3) goes through the intermediate set G before terminating in D. -10- 1 ll ! II lis ill i" Sequence (k°k) goes from A to B, and since B is final this must be its terminal set. This sequence enters a cycle of states in B and does not terminate. Sequence (hz^) remains, indefinitely, in the pseudo-final set A. It may appear as if condition (3? 3b) is violated in the cycle because the fourth signal never changes after the first two states but since a^' (k) = a. (k) for states (0110) and (0010) we see that no violation has occurred „ 5* Speed Independence 5.1 In the previous discussion we have sought to describe the behavior of a circuit by means of the concept of an allowed sequence. In (3»F) we saw that the circuit eventually passes into a final or pseudo-final set of states and stays in this set from then on. If the terminal set consists of one state, this means that the circuit goes to equilibrium and the allowed sequence terminates Otherwise the circuit may never reach equilibrium but one may regard the final or pseudo-final set, which perpetuates itself, as a sort of dynamic equilibrium. 5.2 Of special interest are those circuits in which the ultimate behavior of the circuit does not depend on the relative speeds of the elements . Such circuits, which will be called speed independent, may be designed without regard to time tolerances and capacities of elements and wiring. Hence they should be easier to design and more reliable than asynchronous circuits which require time tolerances on the elements for proper operation. There remains the problem of selecting a suitable definition for speed independence in asynchronous circuits <, One might choose the requirement that all allowed sequences starting with a given initial state u shall terminate in an equilibrium state m. This requirement has been expressed before in slightly different terms by another author , and it has the advantage that it ensures a unique final condition for the circuit „ On the other hand, it precludes the possibility of having speed independent circuits which "cycle" indefinitely (within a final or pseudo-final set) and for this reason we shall use a somewhat broader definition for speed independence „ (5°l) A circuit is said to be speed independent with respect to an initial state u if there exists a final set M such that every allowed sequence beginning with u contains a state in M. ■11- (5:A) If a circuit is speed independent with respect to u, then the terminal set of every allowed sequence starting with u must be M. Proof: Let an allowed sequence under consideration be u, a(l), a(2), ... , and let a ( j ) be a state of the sequence which is in M. Then a(j),a(j+l), ... must all lie in M for if a(k) with k > j is in M*, then M & M*, which is impossible. Thus M is the terminal set of all allowed sequences starting with u. Theorem (5 'A) shows that the ultimate behavior of the circuit, when placed in state u, will not depend upon the relative speeds of the elements . That is, we interpret the rather loose concept of ultimate behavior as meaning a speci- fication of which terminal set is attained by an allowed sequence. Thus if all allowed sequences starting with u have the same terminal set we mean that the circuit will always arrive, ultimately, at a unique static or dynamic condition. (5:B) A circuit is speed independent with respect to u if and only if U, the set of u, is followed by just one final set and no pseudo-final sets . Proof : Assume U <$ M, where M is final, and that U is followed by no other final set and no pseudo-final sets. Then any allowed sequence u, a(l),a(2), ... must have M as its terminal set, for if the terminal set T of the sequence is not equal to M then U & T in contradiction of the hypothesis . Assume the circuit is speed independent with respect to u. Now if U & T, where T is either final or pseudo-final but not equal to M, we may construct an allowed sequence whose initial state is u and whose terminal set is T by (3°G). Since by hypothesis this sequence contains a state m of M, we have M ^ T which contradicts the assumption that M is final. 5-3 When we face the involved problem of circuit design it becomes apparent that (5*1) is not a convenient definition for speed independence from the point of view of applications . It is usually difficult to determine whether or not a given circuit is speed independent since the number of states, sets, and allowed sequences starting with an initial state u may be very great. Therefore, in the following paragraphs we shall discuss other conditions which do not have this disadvantage yet imply speed independence . 1 2 5 '4 One such condition which has been used by other authors ' is: (5'2) A circuit is totally sequential with respect to an initial state u if and only if there is only one allowed sequence starting with u. -12- ll ! H This condition does imply speed independence since U, the set of u is followed by some final set M. Hence, by (3:G) the single allowed sequence starting with u must have terminal set M. If we assume that u,a(l),a(2), ... is the only allowed sequence starting with u then we must have u $a(l), ... ,a ( j ) #£ a(j+l), ... but there can be no other state b(j+l) such that a ( j ) 5^ b(j+l). Thus a(j+l) = a'(j) and by (3:1) we see that there is at most one signal a.(j) of a(j) for which a.(j) 4 a -'(j)* For this signal we must have either a . ( j ) = a . ' ( j ) + 1 or a.(j) = a.'(j)-l. Thus, only one signal tends to change at a time in a totally sequential circuit so that no circuits can be totally sequential and have parallel changes occur in them. Many modern computers use just such parallel changes to achieve their extraordinary speeds. Therefore we seek some other condition to impose on our circuits which is less restrictive than that they be totally sequential. (5:3) A circuit is semi -modular with respect to an initial state u if and only if for all states b and c such that u <£ b and b 0\ c the relationship c 2R b 1 also holds. ^ .k The name "semi -modular" was given to this condition because of its connection with a semi -modular lattice which will be discussed later. An immediate consequence of (5: 3) is the property that for any three states a, b, and c such that u cr a, a £R b and a (K c there must be a fourth state d such that b 3R d and cj^d. We see that d = a' has these properties. A more useful state d of this type is defined by: (5:^-) If a circuit is semi -modular with respect to u and u era, a J a . ' v ' l v \' l l—i This definition is not inconsistent in case a. = a , ' since then a.=b.=c. . l i l l i (5:5) In (5:*0 a lR d, b# d, and c # d. Proof: Since b. and c. both satisfy the conditions of (3:1) we see that d. i i J v ' l must also and hence a?Rd. Now if a . < b . < a . ' we have b .' < d . < a . ' < b ' by l—i.i .1—1—1-1 (5:3) and (3:1). Similarly if a . > b . > a . * we have b. > d . > a . : >b.'. However, -13- if Id . = a. ' then we have b = d. . Hence in any case we satisfy (3:1) giving biRd. Similarly c }R d . (5:6) In a circuit semi -modular with respect to u let u m*. Construct a partial allowed sequence a(j), b(l), b(2), ... ,b(h), m*. (Again let b(0) = a ( j ) and b(h+l) = m* for notational convenience.) By (5^6) we may define c(l) = M[a (j ) ja (j+l) ,b (l)] and c(p) = M[b(p-l); b(p), c(p-l)] for p = 2,3, ...,li+l. Also by (5:5) we have a(j+l)3? c(l) & c(2)# . .. ^ c(h+l). Now we obtain a(j+l)<7 c(h+l) by (3:9b). Also c(h+l) ^ m* since a (j+l),*7 m* and by (5:5) nr* = b(h+l) o" c(h+l). Thus M* cannot be final in contradiction of our hypothesis . Next assume that U is followed by a pBeudo-f inal set T composed of states t(0),t(l), ... ,t(k). Since T is not final there must be some set D =f T such that T & D. Construct a partial allowed sequence t(0), ... ,d, where d is a state in D. Let q(0) be the first state in this sequence which is not in T. Thus, we have t(j)$3 a(0) for some state t(j) in T which directly precedes q in the sequence. We lose no generality by making j = since the numbering of the states in T is arbitrary. Thus we take t(0)$^q(0). -ll+- II II I I Il , ! > I !! |J ii i« Construct a partial allowed sequence t(0), ... ,t(l), ... , ... ,t(k), ... ,t(0) in the manner of the proof of (3:G) so that all states in T appear in the sequence. States in this partial allowed sequence may be renamed p(0),p(l),p(2), . . .,p(r),p(0), where p(o)-=t(0) and all the other k states have been renamed as one or more of the states p(j). Since utfp(O) by (3:C) we may define q(j) = M[p(j-1) ; p(j), q(j-l)] for j = 1,2, ... ,r and q(r+l) = M[p(r) ; p(0), q(r)] by use of (5:6)* By (5:5) we have q(0)$ q(l) $. ... ^q(r+l). By (3:9b) we have q(0) <£q(r+l). Now q(r+l) cannot lie in T, for if so we would have p(0)£R q(0) & q(r+l) and q(0) would also lie in T. Thus p(0) \ q(r+l) so that p(0) and q(r+l) must differ in at least one signal, say p i (0) \ q ± (r+l). This implies p i (j) 4 ( l i (j) f ° r all J = 0,1, ... ,r, for if p.(j) = q.(j) then p. (j+l) = q. (j+l) by (5:^) and hence by induction p. (0) = q. (r+l), which violates our assumption. We assume now that p. (0) < q. (0) . Then it follows that p. (j ) < q. (j ) < p. ' (j ) for all j = 0,1, ... ,r. For assume it is true for j = 0,1, ... ,s. Then since p(s)2R q(s) and p j _ (s ) < q ± (s) < p ± « (s), we have q^s+l) = Max [p i (s+l), q i (s)] . But q. (s+l) 4 P- ( s +l) and p( s +l)y? Q.(s+l) so p. (s+l) < q. (s+l) < p. ' (s+l) . Since V ± U) < Pi' (J) for all j = 0,1, ... , r , we have p.(0) < p.(l) < ... < p. (r) < p. (0). Hence p ± (0) = p. ( j ) < p.' (j) for all j = 0,1, ... , r , so that t. (0) = t.(j) < t.' (j ) for all j = 0,1, ... ,k. This contradicts the assumption (3:17) that T is pseudo- final. If, on the other hand we assume that p. (0) > q. (0) we may make a similar argument with the sense of the inequalities reversed and giving t. (0) = t.(j) > t.'(j) for all j = 0,1, ... ,k. This also contradicts assumption (3:17) • .5.5 The condition (5*-3) for semi -modularity is a convenient one to use, since it is defined with respect to individual transitions which may occur between states and does not refer to allowed sequences and the system of states as a whole. These transitions may be examined individually to see if they satisfy b $ a ' when- ever a ^ b, but no such method seems possible if (5:1) is used. It is true that (5:3) is more restrictive than (5:1) but, in practice, no useful circuits have been found which satisfy the weaker and not the stronger condition. Parallel changes for example, may occur, under (5:3) and common computer circuits such as counters, parallel adders, and shifting registers have been designed subject to it. Although circuit design techniques will be discussed in a later paper we should note that all these techniques require at least the assumption of semii-modularity. -15- Bibliography 1. D. A. Huffman, "The Synthesis of Sequential Switching Circuits", Journal of the Franklin Institute, Vol. 257, Nos . 3 and k (195*0 2. G. H. Mealy, "A Method for Synthesizing Sequential Circuits", Bell Systems Technical Journal, September 1955 , P-P- 10^5-79- -16-