1 / 28

Erik Curtis, Marek Perkowski

A Transformation Based Algorithm for Ternary Reversible Logic Synthesis using Universally Controlled Ternary Gates. Erik Curtis, Marek Perkowski. Synthesis Problems for Cascades. Gates are binary and ternary Toffoli, Feynman, NOT GTG Subset of GTG Universally controlled (binary and ternary).

sue
Download Presentation

Erik Curtis, Marek Perkowski

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. A Transformation Based Algorithm for Ternary Reversible Logic Synthesis using Universally Controlled Ternary Gates Erik Curtis, Marek Perkowski

  2. Synthesis Problems for Cascades • Gates are binary and ternary • Toffoli, Feynman, NOT • GTG • Subset of GTG • Universally controlled (binary and ternary) • Algorithms are searches • From inputs to outputs • From outputs to inputs • Bidirectional • Decomposed

  3. Gates 2-input binary gates NOT V 2-input ternary gates op1 deVos op2 k-input binary gates op3 k-input ternary gates …. …. AND arbitrary arbitrary arbitrary op1 op1 op2 op2 NOT V+ op3 op3

  4. A set of all one-qubit gates for ternary reversible logic

  5. Definition • The Universally Controlled Ternary Gate (UCTG) is a n*n gate where the first n-1 wires are unchanged and wire n is transformed by one of the 6 generalized inverters based on an arbitrary function f of wires 1, 2, …, n-1.

  6. 3.1 The Basic Algorithm • The basic algorithm is a greedy one-pass over the entire function that transforms the output vector to the input vector one bit at a time, until the identity function is found. • The algorithm consists of a special case loop for vector 0 and a general case loop for all other vectors in the function. • See figure 1 for the listing of the pseudo-code of the basic algorithm.

  7. Step 1: Let m = # of inputs If (0)  0, then For j in {0..m-1} { if((0,j) = 1 ) apply transform ‘+1’. if((0,j) = 2 ) apply transform ‘+2’.} Step 2: Let + be the current function; ++ is next function. For i in {0..3m-1} { let p be a ternary vector of i For j in {0..m-1} { if(+(i,j) != p[j] ) { let c2 be a list of all bits in p = 2 except j let c1 be a list of all bits in p = 1 except j t = find_transform(+(i,j), p[j])Transform( j, t, c1, c2) } }} Figure 1. Basic Algorithm

  8. Figure 1. Basic Algorithm function Transform( position, transform, c1, c2) { for i in {0 .. 3m-1} { if( meets_constraints( f +(i), c1, c2 ) ) {f ++( i, position) = apply( transform, f +(i,position) ) } }f+ = f++ }

  9. The first step of the algorithm • The special case loop for (0) is executed first. • In this case, there are no control lines associated with the transforms. • If (0) = 0, then vector 0 is in the correct location and no transformation is needed; therefore, the algorithm continues on to step 2. • In the case where (0)  0, the algorithm then loops over every bit j in vector 0. • Wherever j is equal to 1, an uncontrolled transform ‘+1’ is applied to bit j in every vector. • Wherever j is equal to 2, an uncontrolled transform ‘+2’ is applied to bit j in every vector. • At the end of step 1, (0) = 0. • At no time should this vector change from 0 to another value, now that this value has been locked in.

  10. In the second step, the algorithm looks at every vector i, where 1 i 3m – 1, and every bit j in vector i (a minterm) in turn. • A ternary vector p is created where p is the ternary representation of i. • For each i and j, if the bit of the current function +(i,j) is not equal to the expected bit p[j], then apply the proper transform to all bits in position j, with respect to a set of control vectors c1 and c2. • The control vectors c1 and c2 are used to keep the algorithm from transforming the previously locked bits. • The control vectors are created by putting every bit equal to 2 in +(i) except bit +(i,j) into c2, and every bit equal to 1 in +(i) except bit +(i,j) into c1. • Table 2 shows the logic table for transform selection. The second step of the algorithm

  11. This is given as problem description The principle of the DMM-like algorithms The algorithm build the cascade from outputs to inputs. Yellow shows where the argument values are modified (the controlled variable) Operator 12 applied to A Operator +1 applied to A when B=2 Operator +1 applied to B when A=2 B A

  12. = AB Operator +1 applied to A when B=2 Operator +1 applied to B when A=2 Operator 12 applied to A Operator +2 applied to B Operator +1 applied to A Operator +2 applied to A when B=1 Operator +2 applied to A when B=1 B A

  13. = AB Operator +1 applied to A when B=2 Operator +1 applied to B when A=2 Operator 12 applied to A Operator +2 applied to B Operator +1 applied to A Operator +2 applied to A when B=1 Operator +2 applied to A when B=1 B A

  14. Operator +2 applied to A when B=2 F(A,B) G(A,B) L = AB Operator +2 applied to B when A=2 Operator 12 applied to A Operator +2 applied to B Operator +1 applied to A Operator +1 applied to A when B=1 Gates are created from output to input B F A G H I J K L

  15. Operator +2 applied to A when B=2 J(A,B) K(A,B) L = AB Operator +2 applied to B when A=2 Operator 12 applied to A Operator +2 applied to B Operator +1 applied to A Operator +1 applied to A when B=1 Gates are created from input to output B F A G H I J K L

  16. Step 1: Let m = # of inputs If (0)  0, then For j in {0..m-1} { if((0,j) = 1 ) apply transform ‘+1’. if((0,j) = 2 ) apply transform ‘+2’.} Step 2: Let + be the current function; ++ is next function. For i in {0..3m-1} { let p be a ternary vector of i For j in {0..m-1} { if(+(i,j) != p[j] ) { let c2 be a list of all bits in p = 2 except j let c1 be a list of all bits in p = 1 except j t = find_transform(+(i,j), p[j])Transform( j, t, c1, c2) } }} function Transform( position, transform, c1, c2) { for i in {0 .. 3m-1} { if( meets_constraints( f +(i), c1, c2 ) ) {f ++( i, position) = apply( transform, f +(i,position) ) } }f+ = f++ } Figure 1. Basic Algorithm

  17. Table 3. Input and intermediate variables for basic algorithm Critical minterms are red numbers Sorted minterms are pink rectangles

  18. Fig. 3. UCTG realization of basic algorithm. (a) before transformation, (b) after.

  19. Figure. 5. Compressed result.

  20. Figure 4. Compression Algorithm Step 1: For every uncontrolled gate, find the closest controlled gate on the same wire. Move the uncontrolled gate next to the controlled gate. Every gate that the uncontrolled gate passes over that contains a control line attached to the same wire as the uncontrolled gate, has that control transformed by the uncontrolled gates transform, taking into account the direction of movement of the uncontrolled gate. Step 2: Find controlled gates that are on the same wire and are next to each other. If the transforms of these gates are compatible, then the gates can be merged by merging the control lines and converting the transforms into the new merged transforms. Step 3: Find control gates that are on the same wire and are next to each other. If the control lines of these gates are exclusive, then the gates can be merged by setting one control line set to be output 1 and the other control line set to be output 2. Step 4: Merge the uncontrolled gates. Find the uncontrolled gates and merge them into the control gates they are next to, if it is possible.

  21. Table 4. Compatibility Chart

  22. Table 5. Benchmark Results

  23. Abhinav Agarwal – DATE 2004 • Claims to have the best results of all so far. • Many points uncertain • We have to dig deeper to see if there is really any value there that we do not know. • We used Reed Muller earlier but differently. • Is PPRM better than FPRM, may be because of efficient realization on a computer.

  24. a=a a=a a=a a=a+1 ab c b=b+c+ac b=b+ac b=b b=b c=b+ab+ac c=c+ab+ac c=c+ab c=c a cb 00 01 11 10 001 000 111 010 101 110 011 100 R,Q,P From inputs to outputs Uses gates that are their own inverses F(C,B,A)=[1,0,7,2,3,4,5,6] Uses gates that are their own inverses

More Related