arXiv:hep-ex/9705008v1 13 May 1997 The neurochip Totem: a case study in HEP. S. Dusini a,b , F. Ferrari a,b , I. Lazzizzera a,b,2 , P. Lee d , A. Sartori a,c , A. Sidoti a,b , G. Tecchiolli a,c A. Zorat a,b a INFN - Sezione di Padova, Gruppo Collegato di Trento - Trento - Italy b Universit` a di Trento - Trento - Italy c Istituto per la Ricerca Scientifica e Tecnologica - Trento - Italy d University of Kent - Canterbury - United Kingdom Abstract It is being proved that the neurochip Totem is a viable solution for high quality and real time computational tasks in HEP, including event classification, triggering and signal processing. The architecture of the chip is based on a ”derivative free” algorithm called Reactive Tabu Search (RTS), highly performing even for low pre- cision weights. ISA, VME or PCI boards integrate the chip as a coprocessor in a host computer. This paper presents: 1) the state of the art and the next evolution of the design of Totem; 2) its ability in the Higgs search at LHC as an example. Reference number: 303 Key words: Neural networks, Processors, Computer Arithmetic. PACS: 07.05.Mh 55.40.-e 84.35 1 Introduction Neural networks implemented as VLSI hardware are being considered as good candidates to solve problems of time-critical and high quality performance pattern recognition in High Energy Physics (HEP) [1–3]. The main benefit is speed, because of the massive parallel architecture. A cost is usually a very complex architectural structure, since common algorithms such as backprop- agation, being derivative-based, require high precision computation[4]. To gain significant improvement in this respect, Battiti and Tecchiolli de- vised a ”derivative-free” algorithm in the context of a novel approach to the training problem, which is first transformed into a combinatorial optimiza- tion task, then solved by means of the heuristic method called Reactive Tabu 1 Work supported by Istituto Nazionale di Fisica Nucleare (INFN) 2 Corresponding author: Dipartimento di Fisica, Univ. Trento, I-38050 Povo (TN) Tel:+39-461-881551, fax:+39-461-882014, e-mail: lazi@abacus.science.unitn.it Preprint submitted to Elsevier Preprint 7 February 2008