220 likes | 322 Views
On Reliable Modular Testing with Vulnerable Test Access Mechanisms. Lin Huang, Feng Yuan and Qiang Xu. Purpose. Is on-chip data transmission reliable? What is the solution? Correction Retransmission Hybrid schemes They are helpful in normal functional mode
E N D
On Reliable Modular Testing with Vulnerable Test Access Mechanisms Lin Huang, Feng Yuan and Qiang Xu
Purpose • Is on-chip data transmission reliable? • What is the solution? • Correction • Retransmission • Hybrid schemes • They are helpful in normal functional mode • However, how about modular testing? Cross talk Yield loss IR drop Yield loss and even alpha particle hits …
Agenda • Introduction to Modular Testing • Test Data Transmission “Error-Free” Assumption • Impact of Fault-Tolerant Schemes • The Proposed Solution • “Jitter-Aware” Test Wrapper Design • “Jitter-Transparent” ATE Interface Design • Experimental Results • Conclusion
Introduction to Modular Testing • “Divide and conquer” manner • test wrapper • test access mechanisms (TAMs) • ATE Interface basic • TAM designs classification • dedicated bus-based access scheme • functional access scheme
“Error-Free” Assumption is Questionable • Existing work assumes test data transmission to be error-free • It is questionable when at-speed functional interconnects are reused as TAMs
Fault Tolerance Schemes • Retransmission and hybrid schemes are mainstream techniques to achieve fault-tolerant communication • Retransmission brings problems • Test traffic jitter • Test bandwidth mismatch
The Significance of These Problems • Given the number of flits in the entire test data volume , the flit error rate , the potential yield loss can be expressed as • When and flit size is 32 bits, the test yield loss for the chip containing 21.5M gates [1] is 11.47% ! [1] C. Barnhart et al, Extending OPMISR Beyond 10x Scan Test Efficiency. IEEE Design & Test of Computers, 19(5):65-73, Sep.-Oct. 2002
Buffer-Only Solution • Given the flit injection rate is 0.1 flits per cycleand the extra delay caused by one retransmission is 40 cycles st nd The 1 The 2 retransmission retransmission Shortage of reserved flits 5 reserved flits 1 reserved flits
“Jitter-Aware” Test Wrapper Design • Two extra states: • HALTIN • HALTOUT Input HALT IN Blocked SHIFT CAPTRUE Output Blocked HALT OUT IEEE 1500 Finite State Machine
Wrapper Input Control Kernel Control IPath _ Blocked MCmd _ I n SCmdAccept _ I n S can _ E n IBMU _ Ctrl Block _ Ctrl Output Control SCmdAccept _ O ut Gated _ Clk MCmd _ O ut OPath _ Blocked OBMU _ Ctrl S can _ Clk OCP _ Clk Control Logic Clock Division Control Logic of Test Wrapper
“Jitter-Transparent” ATE Interface • If the ATE operate in a stream mode … • Minimum buffer size: that is able to tolerate the extra delay caused by one retransmission • Given the flit error rate and the number of flits , the test yield loss can be computed as follows: Without Buffer:
“Jitter-Transparent” ATE Interface • We propose to divide the entire input test data flow into segments and insert a small section of “don’t-care” bits Data transmission direction
Test Yield Improvement • Given the flit error rate , the number of flits , and the number of segments , the test yield loss can be computed as follows: Without Buffer: With Minimum Buffer Size:
Experimental Setup • Commercial 90nm CMOS technology • Area overhead • 838 two-input NAND equivalent gates • An industrial circuit[2] • Number of gates: 2.6M • Number of scan cells: 274K • Compressed scan test data volume: 106M • System parameters • Flit injection rate: 0.25 flit per cycle • Flit size: 32 bits • Retransmission delay: 40 cycles [2] C. Barnhart et al, OPMISR: The Foundation for Compressed ATPG Vectors. In Proc. IEEE International Test Conference, pp. 748-757, Nov. 2001
Yield Loss: 0.05% Yield Loss: 4.41% Test Yield Loss for a Core with 2.6M Gates Flit Error Rate: 10-7 Yield Loss: 28.20% Cost: Testing Time Penalty: 0.15%
Proposed Technique vs. Buffer-Only Solution nb: Buffer size for buffer-only solution λ: Flit error rate np: Buffer size for the proposed design ΔTp: Testing time extension ratio of the proposed design
Conclusion • The “error-free” assumption of existing work is questionable • Fault-tolerant schemes may lead to traffic jitter and variable test bandwidth • We propose a “jitter-aware” test wrapper and an on-chip “jitter-transparent” ATE interface to achieve reliable modular testing • Experimental results demonstrate the effectiveness