1. Scalable Test Pattern Generator Design Method for BIST Petr Fišer, Hana Kubátová Czech Technical University in Prague Faculty of Electrical Engineering Dept. of Computer Science & Engineering

2. LATW 2008, 17. – 20.2., Puebla2 Outline Motivation Proposed BIST scheme The Column-Matching algorithm Customizing the design process Experimental results Conclusions

3. LATW 2008, 17. – 20.2., Puebla3 Motivation Four important aspects in BIST: Fault coverage Test time BIST area overhead BIST design time

4. LATW 2008, 17. – 20.2., Puebla4 Motivation Different ASIC designers have different requirements for BIST Design the BIST equipment as fast as possible, regardless the area overhead and the fault coverage (no time to wait, the deadline is approaching!) Design the BIST equipment to be as small as possible, regardless the time it takes (low power) High fault coverage is the most important aspect, the area overhead is next. The design time is not that important (common practice)

5. LATW 2008, 17. – 20.2., Puebla5 Motivation We propose a method offering a big trade-off between all these criteria: Fault coverage Test time BIST area overhead BIST design time

6. LATW 2008, 17. – 20.2., Puebla6 Motivation We propose a method offering a big trade-off between all these criteria: Fault coverage Test time BIST area overhead BIST design time Given by ATPG 100% fault coverage will be considered

7. LATW 2008, 17. – 20.2., Puebla7 The Proposed BIST Mixed-Mode (Hybrid) BIST Two separated phases: pseudorandom and deterministic

8. LATW 2008, 17. – 20.2., Puebla8 The Decoder Design The Column-Matching method Deterministic BIST: LFSR produces pseudo-random code words (C-matrix) These are then transformed into deterministic tests computed by ATPG (T-matrix)

9. LATW 2008, 17. – 20.2., Puebla9 The Decoder Design The Column-Matching method – basic principles Try to reorder test patterns, so that most of the Decoder outputs will be implemented as wires – a Match This will be accomplished when two particular columns of the LFSR and test matrices will be equal Combinational logic – the order is insignificant Unmatched outputs have to by synthesized by a Boolean minimizer

10. LATW 2008, 17. – 20.2., Puebla10 The Decoder Design The Column-Matching method – example y0 = x0 y1 = x1 y2 = x2’ y3 = x1 y4 = x0’ + x1

11. LATW 2008, 17. – 20.2., Puebla11 Mixed-Mode BIST 1.Simulate several LFSR patterns 2.Determine undetected faults 3.Compute a test for them (ATPG) 4.Design a decoder generating vectors for this test and following LFSR patterns

12. LATW 2008, 17. – 20.2., Puebla12 Mixed-Mode BIST Column Matching

13. LATW 2008, 17. – 20.2., Puebla13 Mixed-Mode BIST Decoder Switch

14. LATW 2008, 17. – 20.2., Puebla14 Scaling the Lengths of the Phases Pseudorandom phase To detect easily detectable faults Deterministic phase To generate deterministic vectors Longer PR phaseLonger Det. phase BIST design timeDecreasedIncreased BIST area overheadDecreased BIST run lengthIncreased

15. LATW 2008, 17. – 20.2., Puebla15 Reducing the LFSR Width By weighted pattern testing 3-weight logic The weights are computed for RPRFs The LFSR inputs are AND-ed and OR-ed to produce weights

16. LATW 2008, 17. – 20.2., Puebla16 Reducing the LFSR Width ISCAS’89 s13207.1 benchmark (700 inputs) LFSR (r)w. gatesDesign time [s]GEs 700 (no weights)-47202975 700 (3 weights)5182453361 200 (3 weights)3656531231 50 (3 weights)3652835671 45 (3 weights)3653423693 40 (3 weights)365294341288 30 (3 weights)365--

17. LATW 2008, 17. – 20.2., Puebla17 Experimental Results – Area Overhead Comparison Comparison with state-of-the-art methods. Equal test lengths, the area overhead is compared Compared with: Bit-Fixing [N.A. Touba, E.J. McCluskey: Bit-Fixing in Pseudorandom Sequences for Scan BIST, IEEE Transactions on CAD, Vol. 20, No. 4, April 2001, pp. 545-555] Weighted-pattern BIST [S. Wang: Low Hardware Overhead Scan Based 3-Weight Weighted Random BIST. Proc. 2001 IEEE International Test Conference] Row Matching [M. Chatterjee, D.K. Pradhan: A BIST Pattern Generator Design for Near-Perfect Fault Coverage, IEEE Transactions on Computers, vol. 52, no. 12, December 2003, pp. 1543-1558] CM always better CM better in 71% CM better in 60%

18. LATW 2008, 17. – 20.2., Puebla18 Experimental Results Hard-to-test (and “big”) benchmarks BenchPR testabilityTest lengthTime [s]Overhead c26704.5 M5 K43715 % c7552> 100 M10 K88716 % s9234.110 M200 K35005 % s13207.1100 K50 K130.6 % s15850.1> 10 M100 K12005 % s38417> 10 M100 K460014 % s35841.1> 1 G100 K341 % b125 M10 K10807 % b15> 100 M1 M480017 %

19. LATW 2008, 17. – 20.2., Puebla19 Conclusions The Column-Matching principle proposed Influence of the lengths of the phases is studied Very scalable, many design parameters freely adjustable The results obtained by CM are mostly better (wrt. the area overhead) than those obtained by state-of-the-art methods The method should serve as a basic guideline how to design more complex BIST designs, i.e., the multiple ‑ scan chain based BIST, the STUMPS architecture, etc. It can be very advantageously used to test SoCs, since the LFSR may be reused for more cores.