1ECE 269Krish Chakrabarty1ECE 269VLSI System TestingKrish ChakrabartyLecture 20:Bridging FaultsECE 269Krish Chakrabarty2Outline• Some perspective• Non-feedback bridging faults• Feedback bridging faults• Test generation2ECE 269Krish Chakrabarty3Perspectives and Trends• In the past, bridging faults were seldom considered in practice– Too many– Strength properties ill-defined– May introduce feedback and oscillation• Usually considered as an “add-on” to SSL faults• Today, submicron technology makes it necessary to consider bridging faults directlyECE 269Krish Chakrabarty4Types of Bridging Faults• Non-feedback bridging fault does not create feedback—relatively easy to handle• Feedback bridging fault (FBF) transforms combinational circuit to become sequential• FBF may even cause oscillations3ECE 269Krish Chakrabarty5AND Non-Feedback BFTheorem: Test t detects the AND NFBF between x and y iff t detects x/0 and sets y to 0, or vice versa.(Application: NFBF test generation can be done using ATPG for SSL faults)Proof: Sufficiency (⇒)Suppose t detects NFBF between x and y. Then t makes (x,y) = (1,0)or (x,y) = (0,1). Moreover, t creates a sensitized path to the primaryoutput. If t makes (x,y) = (1,0) then it detects x/0 and places 0 on y. Inthe other case, …Necessity (⇐)Suppose t does not detect the NFBF. Then either it does not activatethe fault (places 00 or 11 on xy) or does not create a sensitized path...(Extend to OR by interchanging 1 and 0)ECE 269Krish Chakrabarty6AND NFBFs• Corollary: Let x and y be inputs to a (N)OR gate, and let x and y not both fan out. Then the AND NFBF between x and y dominates x/0 and y/0.• Conjecture [Friedman, 1974]: The above result holds when both x and y fan out4ECE 269Krish Chakrabarty7AND Feedback Bridging Faults• Theorem: If t detects f/0 and sets b to 0, then t detects F = (b.f)• Theorem: If t detects b/0 and sets f = 0 without sensitizing the path b-f then t detects F = (b.f)bfECE 269Krish Chakrabarty8Test Generation for BFs(1) Create “workaround” circuit containing an SSL equivalent to the BFxyzz*(x.y) ≡ a/0Problems:• Circuit complexity is greatly increased• Unnecessary and meaningless SSL faults are introducedxyzz*a5ECE 269Krish Chakrabarty9Test Generation for BFs(2) Simulation (implicit) using preceding relations between BFs and SSL faultsExperimental Data [Abramovici and Menon, 1985, pre-submicron era]• Good SSL test sets (95% fault coverage) detect most (83%) BFs• Some bad cases exist• High percentage (75%) of potentially oscillating BFs were detected• Overhead for simulating BFs is modest• Nowadays, explicit testing of BFs is necessary (IDDQ
View Full Document
Unlocking...