Automatic synthesis of analog fuzzy controllers

Au oma ic Syn hesis o Analog Fuzzy Con olle s R. G. Ca ajal, A. To alba, F. Colod o and L. G. F anquelo Dp o. de Ingenie ´ıaElec ´ onica EscuelaSupe io de Ingenie os, A da. ReinaMe cedes s/n, Se illa–41012(SPAIN) e–mail: [email p o ec ed] Abs ac —In hispape ,asiliconcompile o analog uzzycon olle s is p esen ed. The layou is au oma ically gene a ed au oma ically, s a ing omase o use speci ica ionssuchasnumbe o ules,num- be o inpu s and maximum cu en (powe consump ion). This ool elies on a lib a y o pa ame izablebasic cells designedin a CMOS 0.7 m echnology.Thes uc u eo hede uzzi ica ionpa o hecon- olle is based on he modula de ini on made in [1]. This s uc u e allows mul i–inpu , mul i–ou pu uzzy logic based con olle s. I also a allows an easy ex ension o he con olle (Se ial o pa allel conec iono hem o ob aincon olle s wi hmo e inpu so ules). I. INTRODUCTION In he las ew yea s, uzzycon olle s ha e ound a la ge numbe o applica ions in ields such as au omo i e indus y, obo ics, powe d i es and home appliances [2]. The e o e, i is no longe s ange o ind uzzy sys emsembedded in indus- ial ASICs. Hence, he a chi ec u e o hese con- olle s is so equen ly epea ed, ha au oma ion is possible. Au oma ion signi ica i ely sho ens he design ime, educing he design cos . Di e - en oolsha e ecen ly been p oposed oau oma e he ha dwa e design o uzzy con olle s using a digi al app oach [3], [4], [5] o an analog app oach [6]. Analog ci cui s a e known o be inhe en ly as e han digi al ci cui s wi h signi ican ly less powe consump ionand silicon a ea. On heo he hand, digi al ci cui s a e supe io in accu acy, ex- endabili y and ease o design. Bu in he bo h cases he design e o is high, so he de elopme o ools o au oma e he design cycle is e y con- enien . In his pape a silicon compile o he au oma ic syn hesis o analog uzzy con olle s is desc ibed. I is based on he modula s uc u e p oposed in [1]. This s uc u e allows o spli in o modules, ha uses a cu en mode app oach. These mod- ules o m a lib a y o pa ame izable basic cells which a e ins an ia ed o build he con olle ha mee s a se o use eque imen s. Once he ne lis isgene a ed, he inallayou can beob ainedusing Cadence So wa e. This pape is o ganized as ollows. Sec ion II de- sc ibes hedesign low. Thea chi ec u eo hecon- olle is de ailed in Sec ion III. Sec ion IV p esen s some esul s and ad ances u u e esea ch. II. DESIGN FLOW S uc u al o analog desc ip ion ci cui Es ima ed Selec ion A chi ec u e Reque imen s Use NO Pe o mances Accep ? YES LAYOUT Accep ? YES NO (Ex e nal o he Compile ) Figu e 1: Design Flow The design lowisdepic ed in igu e 1. In a ypical design, use speci ica ions a e ela ed o numbe o inpu s, numbe o ou pu s, numbe o ules, maximum bias cu en (powe consum ion) and desi ed h oughpu . The silicon compile akes ad an age o he ac ha ha dwa e uzzy con olle s a e, in p ac ice, composed o a se o basic building blocks, such as membe ship gene a o s, max–min blocks and de uzzi ie s. III. ARCHITECTURE OF THE FUZZY COTROLLER The loo -plan o he uzzy con olle is shown in igu e 2. As men ioned be o e i uses he mod- ula s uc u e p oposed in [1]. This s uc u e is based on he uzzy single on algo i hm wi h he Mandani– ype in e ence p ocess and he cen e o g a i y de uzzi ie me hod. As can be seen in he igu e, hiss uc u e allowsa a iable numbe o inpu s, ou pu s and ules. The au oma ic syn hesize conec s hese basic modules (lib a y cells) o o m he comple e con olle . The ollowing building blocks can be iden i ied in he uzzy con olle [7]–[8]: 1. The (MBF) Membe ship Func ion Gene a o . I uses he ci cui p oposed in [9] which is based on a linea unable OTA p oposed in [10]. I p oduces a ully p og ammableMBF’s 1 Iw Iy Ic Is IY IsIwi = Σ ι Ici Iwi Iyi Is =* IsIwi = Σ ι Ici Iwi * = Σ ι Iy I Ic1 s MBF MBF MIN MIN MBFMBF XX 1nIw1 Iy1 CM CM1 PROG PROG One ou pu De uzzi ie Xi(inpu s) (ou pu ) Figu e 2: A chi ec u e o he FuzzyCon olle wi h a T iangula /T apezoidal shape, as can be seen in igu e 3. As he ci cui os based in a linea ansconduc o , i pe o ms he ian- gula and apezoidal shapes be e han he commonlyusedOTAbasedon hedi e en ial pai . 2. The In e ence Machine (IM). The IM blocks pe o ms heMamdani– ypeo in e ence p o- cess( igu e4)using heANDope a o s,which a e implemen ed wi h a MIN ci cui . Figu e µR w1 Rule 1 Rule 2 MIN z z 0 x21 0 x w2 In e ed Fuzzy Se Figu e 4: Mamdani– ype in e ence p ocess, using single on as membe - ship unc ions o he ou pu a iable I MBF2 I MBFK I MBFmax I MBF1 V Vdd bias Figu e 5: Max Ci cui y 5 shows he MAX ci cui , which is based on he Winne –Take–All ci cui p oposed in [11] and la e imp o ed in ([12]); he only modi- ica ion included he e is a cascode ou pu o imp o e he esponse o he ou pu cu en mi o s. This ci cui also buil s he MIN op- e a o , using he De Mo gan’s Law and he complemen ope a o [13]. (Figu e 6) I MBF1 I amp I MBF1 I amp MBFmin I Vbias I MBF1 I MBF MAX CIRCUIT I MBFmax Figu e 6: Min Ci cui y 3. The De uzzi ie Block (DF). The de uzzi ie block implemen es he cen e o he g a i y me hod wi h single ons as he ule conse- quen s. Equa ion (1) esumes he compu a- ions o be pe o med. ou pu = P i = i = 1 w i c i P i = i = 1 w i ( 1 ) whe e is he numbe o ules, w i is he in- e ed ou pu o ule i , and c i is he single on o he consequen o ule i . These compu a- ionscanbepe o medbe e incu en -mode (inaci cui node heou pu cu en is hesum o heinpu cu en s), using hecu en mul i- plie /di ide p oposedin[14]anddepic edin igu e 7. I makes he De uzzi ie Block as e and simple. IV. RESULTS AND FUTURE RESEARCH HSPICE simula ions using a s anda d 0.7 m CMOS echnology we e pe o med. In igu e 10, 9 and 8 can be seen he unabili y o he posi ions , 2 I MBF Vpos1 Vpos2 I MBF Vslop2 Vslop1 Vpos2 Vpos1 IN V I amp I amp Vslop1 Vslop2 I amp I MBF1 I VIN I1 V slop2 I2 Vpos2 I= 1 -I 2 MIN Ci cui MBF2 IMBF2 Figu e 3: Membe ship Func ion Gene a o b I x I y I x I x I y I =* 2*I b ou I Vdd Figu e 7: Cu en mode Mul iplie /Di ide slope and shapes o he membe ship unc ions. In igu e 11 can be seen he ansien ansien o he MIN ci cui . Finally in igu e 13 is showed he DC pe o manceo hecu en mul iplie andin igu e ?? is showed he ansien esponse o his mul i- plie . P esen ly a lib a y o cells is being designed using Cadenca So wa e. V. ACKNOWLEDGMENTS The au ho s would like o acknowledge inancial suppo byCICYT h ough heTIC86-0860p ojec . Figu e 8: Tunabili y o he posi ion o he MBF REFERENCES [1] J. Rami ez–Angulo, K. T eece, P. And ews and T. Choi. “Cu en –Mode and Volage–Mode VLSI uzzy p ocesso a chi ec u e”. P oc. In . Con . on Ci cui s and Sys ems, ISCAS’95,pp 1156–1159, Sea le 1995. [2] Special issue on “Enginee ing Applica ions o Fuzzy Logic”. P oc. o he IEEE, Ma ch 1995. [3] A.Cos a, A.de Glo ia, P.Fa aboshi, and A.Pagni. “A ool o au oma ic syn hesis o uzzycon olle s”, in P oc. o heIEEE In . Con . onFuzzySys ems, O lando1994,pp.1771–1775. 3 Figu e 9: Tunabili y o he slope o he MBF Figu e 10: Tunabili y o he wi h o he MBF [4] T. Holls ein, S. K. Halgamuge and M. Glesne . “Compu e -Aided Design o Fuzzy Sys ems Based on Gene ic VHDL Speci ica ions”.IEEE T ans.onFuzzySys ems, ol.4no.4pp.403–417, No embe o 1996. [5] R.G.Ca ajal,A.To albaandL.G.F anquelo. “AFAN, A Tool o he Au oma ic Syn hesis o Neu al and Fuzzy Con olle s wi h A chi- ec u e Op imiza ion”, 1997 IEEE In e na ional Symposium on Ci cui s and Sys ems, ISCAS’97. June 9-12, Hong-Kong [6] N. Mana esi, R. Ro a i, E. F anchi, R. Gue - ie i and G. Bacca ani. “A Silicon Complile o Analog FuzzyCon olle s: F omBeha io al Speci ica ions o Layou ”.IEEE T ans. on Fuzzy Sys ems, ol. 4 no. 4 pp. 418–428, No embe o 1996. [7] C.C.Lee. “Fuzzy logic in con ol sys ems: Fuzzylogiccon olle ,Pa I”.IEEET ans.Sys ., Man, and Cybe n., ol. 20, no. 2, pp. 404–418, 1990. Figu e 11: T ansien analisis o he MIN ci cui Figu e 12: DC pe o mance o he mul iplie ci cui [8] C.C.Lee. “Fuzzy logic in con ol sys ems: Fuzzy logic con olle , Pa II”. IEEE T ans. Sys ., Man, and Cybe n., ol. 20, no. 2, pp. 419– 435, 1990. [9] R. G. Ca ajal, A. To alba, F. Colod o and L. G. F anquelo. “An Analog CMOS Uni e - salMembe shipFunc ionGene a o wi hFully Independen Adjus able Pa ame e s”. submi - ed o publica ion [10] S. C. Huang and M. Ismail, “Linea unable COMFET ansconduc o s”, Elec onics Le e s, ol 29, pp. 459-461, Ma ch 1993. [11] J. Lazza o, S. Ryckebusch, M. A. Mahowald and C. A. Mead. “Winne –Take–All Ne - wo ks o O(n) Complexi y”. Ad ances in Neu- al In o ma ion P ocessing Sys ems, ol. 1, D.S. Tou ez ky, ed. Mo gan Hau mann, 1989. [12] Bin-Da Liu and Chun-Yueh Huang. “A - ay Based Fuzzy In e ence Mechanism Im- plemen ed wi h Cu en –Mode CMOS Ci - cui s”.P oc. In . Con . on Ci cui s and Sys ems, ISCAS’94 ol. 5, London, June 1994. 4 Figu e 13: T ansien esponse o he mul iplie ci cui [13] M. Sasaki, T. Inoue, Y. Shi ai and F. Ueno. “Fuzzy Mul iple–Inpu Maximum and Mini- mumCi cui sinCu en ModeandThei Anal- yses Using Bounded-Di e ence Equa ions”. IEEE T ansac ions on Compu e s, ol. 39, no. 6, June 1990. [14] K. Bul and H. Wallinga. “A Class o Ana- log CMOS Ci cui s Based on he Squa e–Law Cha ac e is ic o an MOS T ansis o in Sa u- a ion”.IEEE Jou nal o Solid–S a e Ci cui s, ol. SC-22, no. 3, pp. 357–365,June 1987. 5