210 likes | 354 Views
Supply Voltage Noise Aware ATPG for Transition Delay Faults. Nisar Ahmed and M. Tehranipoor University of Connecticut Vinay Jayaram Texas Instruments, TX. Overview. Objective Prior Work Statistical IR-drop Analysis Power Model Switching Cycle Average Power (SCAP) SCAP Calculator
E N D
Supply Voltage Noise Aware ATPG for Transition Delay Faults Nisar Ahmed and M. Tehranipoor University of Connecticut Vinay Jayaram Texas Instruments, TX
Overview • Objective • Prior Work • Statistical IR-drop Analysis • Power Model • Switching Cycle Average Power (SCAP) • SCAP Calculator • Pattern Generation Framework • Experimental Results • Conclusions
How to generate IR-drop tolerant patterns without significantly increasing number of patterns? Objective • High switching activity during test compared to functional operation • Increased sensitivity of VDSM designs to supply noise • IR-drop during at-speed test: A BIG CONCERN • Present ATPG tools are supply noise unaware • Targets as many faults as possible per pattern • Random filling of X’s during ATPG • Increases switching activity of test patterns
IR-drop Effects? • Delay Failure • An excessive IR-drop can increase the delay of targeted paths • Logic Failure • An excessive IR-drop can significantly reduce the voltage reaching a device -- may function unpredictably • Question: • What is the impact of IR-drop on long and short paths • Answer: • Slack of a path and number of switching define how tolerant a path is to IR-drop
Prior Work • IR-drop issue during at-speed test [Saxena, ITC03] • Static Verification of Test Vectors for IR-drop Failure [Kokrady, ICCAD03] • Power Supply Noise Analysis in Test Compaction [Wang, ITC05] • Preferred-Fill [Remersaro, IEEE D&T 07] • Low-capture TDF pattern generation [Wen, VTS05, VTS06] • Faster-than-at-speed test considering IR-drop [Ahmed, ICCAD03]
Contributions • Novel Power model to measure the average power during at-speed test • Called Switching cycle average power (SCAP) • Consider both the length of paths affected by each pattern and the number of transitions • Pattern generation procedure: • Set SCAP threshold • Perform TDF pattern generation • Identify high IR-drop patterns using SCAP • Replace them with IR-drop patterns
Case Study • ITC’99 benchmark (b19) • 200K gates, 6642 scan cells • 8 scan chains • 4 VDD (VSS) pads • 0.18 nm technology • Frequency = 142MHz • Power rings Width = 20um • Stripes Width = 10um Power/Ground Distribution Network
Statistical IR-drop analysis • Vector-less approach to estimate IR-drop analysis • Assumptions: • Uniform activity over the entire design region • Switching time frame = clock period • 20% Net toggle activity • 2.8% voltage drop in VDD network • 4.5% voltage drop in VSS network • Underestimates average functional IR-drop and power • Non-uniform switching activity • Most activity occurs during early cycle period IR-drop inversely proportional to switching time-frame window
Statistical IR-drop analysis (cont.) What is an average switching time-frame window to estimate average functional IR-drop and power ? • Procedure: • Generate transition fault test patterns • Measure time-span of all switching activity during launch-to-capture window • Average switching time frame = Half cycle period Functional power threshold to identify high IR-drop test patterns
Clock Network Switching ATE clock T FF clock STW (P2) Scan Enable STW (P1) Power Model (SCAP) • Average IR-drop directly relates to average power • Cycle average power (CAP) • Power measured over entire cycle period • Switching cycle average power model (SCAP) • Measured over switching time frame window (STW) CAP = ∑(Ci * VDD2)T SCAP = ∑(Ci * VDD2)STW
Power Model (SCAP) (cont.) • Pattern P1 and P2 • Almost same switching activity • Different switching time frame window (STW) • P2 has smaller STW • SCAP(P2)> SCAP(P1) IR-drop effects on VDD and VSS during pattern P1 and P2 application within 7ns capture window.
Power Model (SCAP) (cont.) • SCAP(P2)> SCAP(P1) ITC’99 benchmark (b19) P2 P1 Switching time frame window (STW) is an important parameter
SCAP Model Validation Another example: Cadence SOC Design (Turbo Eagle) P2 P1 SCAP is a good power model to represent average IR-drop
Test Patterns Design (.v) Physical Design (DEF) SDF SCAP Calculator STAR-RXCT PLI VCS Pattern Power Profile Instance Capacitance extractor Parasitics (SPEF) SCAP Calculator • PLI routine to measure power during launch-to-capture window • Avoids huge VCD file generation for large designs • SCAP = ∑(Ci * VDD)2 STW SDF – Standard delay format (Timing Information) VCS – Gate level simulator (Synopsys) PLI – Programmable language Interface STAR-RXCT – Extraction tool (Synopsys)
Commercial ATPG tool FS ATPG Statistical IR-drop Analysis Fault List Pattern Set SCAP Calculator Thr Yes If SCAP>Thr ? Short-listed Patterns No Exit Pattern Generation Framework With X-fill options Pattern Generation Procedure: Step 1: Run ATPG for all faults Step 2: Exclude high IR-drop Patterns (short-listed patterns) Step 3: Fault Simulate for Short-listed patterns fault list Step 4: Run ATPG for this fault list
Experimental Results SCAP threshold = 20% toggle activity over avg. switching time frame window • Conventional transition fault pattern set • Random fill option for don’t-care bits • 2360 patterns generated VSS network VDD network 860 patterns with SCAP value above threshold
Results (cont.) • New patterns generated with fill-0 and fill-1 options for faults exclusively detected by short-listed patterns • Fill-0 generates 957 low-power patterns instead of 860 patterns with high SCAP value • 97 extra patterns but significantly reduces the SCAP value Fill-0 option
Results (cont.) Fill-1 option Fill Adjacent
Results (cont.) • Comparison of conventional ATPG and the new pattern generation procedure • Slight increase in pattern volume • 4% increase in number of patterns
Conclusion • New pattern generation procedure for IR-drop tolerant pattern generation • Switching cycle average power (SCAP) model for identifying high IR-drop patterns • Considers both switching capacitance and time frame of activity • PLI based SCAP calculator • Efficient way to measure SCAP during launch-to-capture window