Internatonal Journal of Computer Applcatons Technology and Research A New Desgn of Multpler usng Modfed Booth Algorthm and Reversble Gate Logc K.Nagarjun Department of ECE Vardhaman College of Engneerng, Hyderabad, Inda S.Srnvas Department of ECE Vardhaman College of Engneerng, Hyderabad, Inda Abstract: In ths paper we propose a new concept for multplcaton by usng modfed booth algorthm and reversble logc functons. Modfed booth algorthm produces less delay compare to normal multplcaton process. Modfed booth algorthm reduces the number partal products whch wll reduces maxmum delay count a the output. by combnng modfed booth algorthm wth reversble gate logc t wll produces further less delay compare to all other. In the past years reversble logc functons has developed as an mportant research area. Implementng reversble logc has the advantage of reducng the gate count, garbage outputs as well as constant nputs. Addton subtracton operatons are realzed usng reversble DKG gate. Ths modfed booth algorthm wth reversble gate logc are syntheszed and smulated by usng Xlnx 13.2 ISE smulator. Keywords: MBE, Reversble logc gate, DKG, reversble logc crcuts. Carry save adder tree. 1. INTRODUCTION The Contnuous advances of the mcroelectronc technologes makes better use of nput energy, to encode the data more effcently, to transmt the nformaton more faster and relable, etc. In Partcular, many of these technologes handle low power consumpton to meet the requrements of varous outboard applcatons. In these applcatons, a multpler s a fundamental arthmetc unt and used n a great extent n crcuts. The fastness of multplcaton and addton arthmetc s decdes the executon speed and performance of the total calculaton. Because the multpler needs the longest delay wthn the basc operatonal blocks n dgtal systems, the crtcal path s evaluated by multpler n general. For hgh speed multplcaton, the modfed radx-4 booth s algorthm (MBA)[1] s generally used however ths cannot completely solve the problem. In general, the multpler uses Booth s algorthm and array of full adders (FAs)[3], or Wallace tree rather than the array of FAs.,.e., ths multpler prmarly conssts of the three parts: Booth encoder[6], a tree to compact the partal products such as Wallace tree[1], and the fnal adder[1]. Because Wallace tree s to add the partal products. The most effcent way to gan the fastness of a multpler s to cut down the number of the partal products as multplcaton proceeds a seres of addtons for the partal products. To cut down the number of calculaton steps for the partal products, MBA[1] algorthm has been employed mostly where Wallace tree has taken the role of ncreasng the fastness to add the partal products. To ncrease the fastness of the MBA algorthm, many parallel multplcaton archtectures have been explored and they have employed to varous dgtal flterng calculatons. here In ths desgn we use reversble logc gates[2] n the place of full adders to reduce the power and delay. A reversble logc crcut should have features lke use mnmum number of reversble gates[2], use mnmum number of garbage outputs, use mnmum constant nputs. Fg 1: Basc arthmetc steps for multplcaton and accumulaton In ths paper, A New Desgn of Multpler usng Modfed Booth Algorthm And Reversble Gate Logc are mplemented.secton2 dscusses overvew of MAC.Secton3 ntroduces reversble logc gate functon Secton4 covers the proposed MAC and archtecture of CSA tree. Expermental results are showng the smulaton results of proposed desgn. 2. OVERVIEW OF MAC In ths secton, general basc MAC[1] operaton s ntroduced. A multpler can be dvded nto three functonal steps. The frst s radx-2 modfed Booth encodng[1] n whch a partal product s produced from the multplcand and the multpler. second s the adder array or partal product compresson to www.jcat.com 743
Internatonal Journal of Computer Applcatons Technology and Research add all the partal products and change them nto the form of sum and carry. The last step s the fnal addton n whch the fnal multplcaton result s produced by addng the sum and carry. If the process to accumulate the multpled results s ncluded, a MAC conssts of four steps, as shown n Fg. 1, whch shows the operatonal steps. Fg 2: Hardware archtecture of general MAC A Basc hardware archtecture of ths MAC s shown n Fg. 2. It carres out the multplcaton operaton by multplyng the nput multpler X and the multplcand Y. Ths result s added to the prevous multplcaton result Z as accumulaton step. Fg. 2. Hardware archtecture of the general MAC. The - bt 2 s complement bnary number can be expressed as N 2 N 1 2 N 1 2 0 X x x x 0,1.. (1) If the equaton (1) s expressed n base-4 type redundant sgn dgt form to apply the radx-2 Booth s algorthm, t would be. N /21 X 0 d 4 21 2 21... (2) d 2x x x.. (3) If the equaton (2) s used, multplcaton can be expressed as N / 21 2 X Y d 2 Y 0..(4) If these equatons are used, the before mentoned multplcaton accumulaton results can be expressed as N /21 2N 1 0 j0.(5) P X Y Z d 2 Y z 2. Each of two terms on the rght sde of (5) s evaluated ndependently and the fnal result s produced by addng the two results. The MAC archtecture enforced by (5) s called the standard desgn. If the N-bt data are multpled, the number of generated partal products s proportonal to N. In order to add that partal products serally, the tme of executon s also proportonal to N. The fastest archtecture of a multpler, whch uses radx-2 Booth encodng to generate partal products and a Wallace tree based on CSA as the adder array to add the partal products. If radx-2 Booth encodng s used, the number of partal products,.e., the nputs to the Wallace tree, s reduced to half, resultng n the decrease n CSA tree step. In addton, the sgned multplcaton based on 2 s complement numbers s also possble. Due to these reasons, most current used multplers adopt the Booth encodng. 3.Reversble Logc DKG Gate Reversble functon s the man objectve of the reversble logc theory. A 4* 4 reversble logc DKG[2] gate that can work unquely as a reversble Full adder and a reversble Full subtractor s shown n Fg 3. It can be verfed that nput pattern representng to a partcular output pattern can be unquely determned. If the nput A=0, the proposed gate acts as a reversble Full adder DKG gate, and f nput A=1, then t acts as a reversble Full subtractor DKG gate. It has been proved that at least the two garbage outputs are requred by the reversble full-adder crcut to make the output combnatons unque. P B Fg 3: Reversble DKG gate Q AC A D R ( A B)( C D) CD S B C D The bnary full adder/subtractor s capable of handlng the one bt of each nput along wth a carry n/borrow n generated as a carry out/ borrow from addton of prevous lower order bt poston. n bnary full adders/subtractors are cascaded to add two bnary numbers each of n bts. P A g1 Q B g2 R A( B C) BC carry S A B C sum Fg 4: DKG gate mplemented as Full adder P A g1 _ Q C g2 R A( B C) BC Barrow S A B C Dfference Fg 5: DKG gate mplemented as Full subtractor A Parallel adder/subtractor s an cascaded of full adders/subtractors and nputs are smultaneously appled. The www.jcat.com 744
Internatonal Journal of Computer Applcatons Technology and Research carry/borrow produced at a stage s propagated to the next stage. When the control nput A=0, the crcut behaves as a parallel adder, generates a 4 bt sum and a carry out, as shown n Fg 4. If the control nput A=1, the crcut behaves as a parallel subtractor, generates a 4 bt dfference and a borrow out, as shown n Fg 5. Fg 7: Hardware archtecture of the proposed MAC Fg 6: Reversble Parallel adder when A/S=0 4. PROPOSED ARCHITECTURE In ths secton, By examnng the varous archtectures, we have observed that delay can be mproved by usng hgher radx MBA whch reduces number of partal product rows that eventually reduces number of multplcaton thereby mprovng speed. Thus we propose a new hgh speed and area effcent MAC archtectures whch wll be an mprovement over the exstng Archtecture by replacng Radx-2 wth Radx-4 and Modfed Booth Encoder n the Multplcaton Stage. The output of multplcaton and accumulaton stages wll be combned usng hybrd reducton through CSA, CLA and HA whch enhancng speed and effcency. In addton, the Fnal Stage of CSA wll nclude a New Reversble logc DKG gate was replaced a full adder. The new adder wll be an mprovement over the exstng CSA. by replacng Reversble logc DKG[2] gate n CSA The desgn wll be mplemented usng Verlog language and smulated usng Xlnx ISE13.2 Smulator. We expect the proposed MAC wll be useful n hgh performance sgnal processng system. In Fg. 7 the proposed archtecture were shown and t contan booth encoder and CSA & Accumulator tree and the tree contans half adder reversble logc DKG gate whch wll reduce the delay of the adder and a fnal adder to add the fnal sum and carry. The archtecture of the hybrd-type CSA that comples wth the operaton of the proposed MAC s shown n Fg.8, whch performs 8X8-bt operaton. It was formed based on the prevous archtectures. In Fg.8, S s to smplfy the sgn expanson and N s to compensate 1 s complement number nto 2 s complement number. S[] and C[] and correspond to the th bt of the feedback sum and carry. Z[] s the th bt of the sum of the lower bts for each partal product that were added n advance and Z[]s the prevous result. In addton corresponds to the th bt of the I th partal product. Snce the multpler s for 8 bts, totally four partal products(p 0 [7:0]~P 3 [7:0]) are generated from the Booth encoder., Ths CSA requres at least four rows of DKG full adders for the four partal products. Thus, totally fve DKG full adder rows are necessary snce one more level of rows are needed for accumulaton. For an 8-bt MAC operaton, the level of CSA s (n/2+1). The whte square n Fg. 8 represents an reversble DKG full adder and the gray square s a half adder (HA). The rectangular symbol wth fve nputs s a 2-bt CLA wth a carry nput. The crtcal path n ths CSA s determned by the 2-bt CLA. It s also possble to use FAs to mplement the CSA wthout CLA. However, f the lower bts of the prevously generated partal product are not processed n advance by the CLAs, the number of bts for the fnal adder wll ncrease. When the entre multpler or MAC s consdered, t degrades the performance. the characterstcs of the proposed CSA archtecture have been summarzed and brefly compared wth other archtectures. For the number system, the proposed CSA uses 1 s complement, but ours uses a modfed CSA array wthout sgn extenson. The bggest dfference between ours and the others s the type of values that s fed back for accumulaton. Ours has the smallest number of nputs to the fnal adder. In the carry save adder archtecture more number of full adders were used and that wll replaced by reversble DKG gate n the proposed system by ths the overall performance was further ncreased the half adders n carry save adder were used as t s because no bgger dfference n the normal half adder and reversble DKG half adder. In addton, we compared the proposed archtecture wth that the prevous booth multpler. Because of the dffcultes n comparng other factors, only delay s compared. The szes of both MACs were 8X8 bts and mplemented by usng Xlnx 13.2 smulator. The delay of ours was 30.38ns whle n prevous t was 34.35ns,whch means that ours mproved about 4.2ns of the delay performance. Ths mprovement s manly due to the reversble logc DKG full adder. The archtecture from the prevous should nclude a normal full adders that wll consumes maxmum delay for that we used s very effectve and gves maxmum performance. www.jcat.com 745
Internatonal Journal of Computer Applcatons Technology and Research S0 C[1] P0[7] C[0] Z[6] P0[7] Z[6] P0[6] Z[5] P0 [5] Z[4] P0[4] Z[3] P0[3] Z[2] P0[2] Z[1] P0 [1] Z[0] P0[0] 1 0 S1 C[3] C[2] N0 P1[7] P1[7] P1[6] P1[5] P1[4] P1[3] P1[2] P1[1] P1[0] Z[1:0] S2 C[5] C[4] N1 P1[7] P2[7] P2[6] P2[5] P2[4] P2[3] P2[2] P2[1] P2[0] Z[3:2] S3 C[7] C[6] N2 P3[7] P3[7] P3[6] P3[5] P3[4] P3[3] P3[2] P3[1] P3[0] Z[5:4] N3 S[7] S[6] S[5] S[4] S[3] S[2] S[1] S[0] Z[7:6] C[7] S[7] C[6] S[6] C[5] S[5] C[4] S [4] C[3] S[3] C[2] S[2] C[1] S[1] C[0] S[0] Fg 8: Hardware archtecture of proposed CSA tree 5. EXPERIMENTAL RESULTS The proposed archtecture was smulated and syntheszed by usng Xlnx 13.2 tool and cadence vrtuoso Fg 9: TOP Level Schematc of proposed MAC www.jcat.com 746
Internatonal Journal of Computer Applcatons Technology and Research Fg 10: Output of Modfed booth algorthm wth reversble gate logc Table 1.Comparson table of dfferent parameters Parameter MBE [1] MBE wth reversble gate logc No. of slces used 178 out of 4656 (3%) 162 out of 4656 (3%) No. of LUT s 355 out of 9312(3%) 294 out of 9312 (3%) Total Delay 34.535ns 30.38ns 6.CONCLUSION In ths paper, a new MAC archtecture to perform the multplcaton-accumulaton, To effcently process the dgtal sgnal processng and multmeda applcaton ths archtecture was proposed. the overall MAC performance has been mproved By elmnatng the ndependent accumulaton process that has the greatest delay and mergng t to the compresson process of the partal products, almost twce as much as n the prevous work and by replacng full adder n the CSA wth reversble logc gate further mproves the performance. The proposed hardware was mplemented and syntheszed through Xlnx ISE 13.2 tool. Consequently, the proposed archtecture can be used effectvely where we requrng hgh throughput such as a real-tme dgtal sgnal processng. [3] O. L. MacSorley, Hgh speed arthmetc n bnary computers, Proc.IRE, vol. 49, pp. 67 91, Jan. 1961. [4] S. Weser and M. J. Flynn, Introducton to Arthmetc for Dgtal Systems Desgners. New York: Holt, Rnehart and Wnston, 1982. [5] A.R. Omond,Computer Arthmetc Systems. Englewood Clffs, NJ:Prentce-Hall, 1994. [6] A.D.Booth, A sgned bnary multplcaton technque, Quart. J.Math., vol. IV, pp. 236 240, 1952. [7] C. S. Wallace, A suggeston for a fast multpler, IEEE Trans. Electron Computer., vol. EC-13, no. 1, pp. 14 17, Feb. 1964. [8] A.R. Cooper, Parallel archtecture modfed booth multpler, Proc. Inst. Electr Engg. Vol. 135, pp 1251-1988. 7.REFERENCES [1] A New VLSI Archtecture of Parallel Multpler Accumulator Based on Radx-2 Modfed Booth Algorthm Young-Ho Seo, Member,IEEE,and Dong-Wook Km, Member, IEEE [2] A Dstngush Between Reversble And Conventonal Logc Gates B.Raghu Kanth, B.Mural Krshna, M. Srdhar, V.G. Shant Swaroop, www.jcat.com 747