DocumentCode
2545024
Title
Symbolic Model Checking of Analog/Mixed-Signal Circuits
Author
Walter, David ; Little, Scott ; Seegmiller, Nicholas ; Myers, Chris J. ; Yoneda, Tomohiro
Author_Institution
Utah Univ., Salt Lake, UT
fYear
2007
fDate
23-26 Jan. 2007
Firstpage
316
Lastpage
323
Abstract
This paper presents a Boolean based symbolic model checking algorithm for the verification of analog/mixed-signal (AMS) circuits. The systems are modeled in VHDL-AMS, a hardware description language for AMS circuits. The VHDL-AMS description is compiled into labeled hybrid Petri nets (LH-PNs) in which analog values are modeled as continuous variables that can change at rates in a bounded range and digital values are modeled using Boolean signals. System properties are specified as temporal logic formulas using timed CTL (TCTL). The verification proceeds over the structure of the formula and maps separation predicates to Boolean variables. The state space is thus represented as a Boolean function using a binary decision diagram (BDD) and the verification algorithm relies on the efficient use of BDD operations.
Keywords
Boolean functions; Petri nets; binary decision diagrams; hardware description languages; integrated circuit modelling; logic design; mixed analogue-digital integrated circuits; temporal logic; Boolean based symbolic model checking algorithm; Boolean function; Boolean signals; Boolean variables; VHDL-AMS description; analog/mixed-signal circuits; binary decision diagram; hardware description language; labeled hybrid Petri nets; temporal logic formulas; timed CTL; Analog circuits; Binary decision diagrams; Boolean functions; Circuit simulation; Cities and towns; Formal verification; Logic; Oscillators; Real time systems; State-space methods;
fLanguage
English
Publisher
ieee
Conference_Titel
Design Automation Conference, 2007. ASP-DAC '07. Asia and South Pacific
Conference_Location
Yokohama
Print_ISBN
1-4244-0629-3
Electronic_ISBN
1-4244-0630-7
Type
conf
DOI
10.1109/ASPDAC.2007.358005
Filename
4196051
Link To Document