A Scan Architecture For High Test Coverage & Low Power Consumption
7th July 2015Binod KumarM.Tech(EE)
Outline
Two Conflicting Goals in the Scan technique Introduction to California Scan
Architecture(CSA) Working of CSA Results of CSA Proposal of a new architecture Proof of Suitablity of the new architecture
Two Conflicting Goals in the Scan technique
Most of the bits in test patterns generated by automatic test pattern generation (ATPG) tools are dont-care bits.
To reduce power consumption during scan shift-in operation, dont-care bits can be assigned using the repeat-fill technique.
To enhance defect coverage, dont-care bits can be assigned using the random-fill technique. But this approach increases overall power consumption during the scan shift-in operation.
CALIFORNIA SCAN ARCHITECTURE(CSA)
Proposed by Kyoung Youn Cho, Subhasish Mitra and Edward J. McCluskey in ITC 2007.
CSA provides a trade-off between test quality and power consumption by modifying test patterns during the scan shift-in operation.
Its benefits are
(1) reduced power consumption during test pattern scan shift-in and
(2) improved defect coverage. These benefits are made possible by applying a modified version,
rather than an exact copy, of the scan shift-in pattern to the combinational logic.
CALIFORNIA SCAN ARCHITECTURE(CSA)
Types of Pattern Filling
Repeat Filling-assigns the last care bit 0 filling-fills with all zeroes 1 filling-fills with all ones Random Filling-fills randomly Toggle Filling-assigns dont-care bits using logic-0
and logic-1 alternately. These assignments can improve defect coverage while they increase switching activity during the scan shift-in operation
Illustration of Types of Pattern Filling
Technique ATPG provided pattern Filled Pattern
0 filling 10xxxxxx1xx1 100000001001
1 filling 10xxxxxx1xx1 101111111111
Repeat filling 10xxxxxx1xx1 100000001111
Random filling 10xxxxxx1xx1 100100101001
Toggle filling 10xxxxxx1xx1 101010101101
Working of the CSA
In CSA, the test patterns that are shifted into scan chains have their dont-care bits assigned using the repeat-fill technique.The patterns applied to the combinational logic are altered in the test application process so that they correspond to toggle-fill patterns.
This modification of patterns enhances the defect coverage of test patterns.
CSA operation
States of scan cells during scan shift-in
Pattern to be applied to the combinational logic is 01101010 wich is fairly random & hence not affecting test coverage while scan-in pattern is 11000000 which has very less switching activity.
Experimental Results Of CSA
Authors have compared the results of their own architecture with the following architectures:
1.Traditional Scan architecture with repeat filling
(TSA_rpt)
2.Traditional Scan architecture with random filling
(TSA_rnd)
In results,their architecture is represented as CSA_rpt since CSA also employs repeat filling.
Experimental Results Of CSA (b19)
Experimental Results for SSF test sets
Experimental Results-->N-detect coverage: SSF test sets
CSA:It's not as good as it seems
Let's consider that we need to apply the pattern 0000000 to the combinational logic.
By CSA,101010 has to be scanned in resulting in a larger switching activity,instead scanning in 0000000 itself is a wise option.
Cell6 cell5 cell4 cell3 cell2 cell1
0 1 ? ? ? ? ?
1 0 0 ? ? ? ?
0 1 1 1 ? ? ?
1 0 0 0 0 ? ?
0 1 1 1 1 1 ?
1 0 0 0 0 0 0
CSA:It's not as good as it seems
CSA pays the penalty of higher scan-in shift power for those patterns which have no (or very less) bit transitions in them because even for them it needs a toggled pattern to be scanned in.
A mechanism which decides in an intelligent way for what is to be scanned in after seeing the pattern to be applied.
Proposed new architecture
Idea is to achieve the low switching activity of CSA_rnd and high test coverage of TSA_rnd.
We propose to employ XOR instead of the inverter employed in CSA.
Proposed new architecture
XOR operation is like: c=(not a)b + a(not b) Here, a=scanbit_applied,b=select_xor,c=scan_in When b=0,c=a (BUFFER operation)
scan_in is same as scanbit_applied When b=1,c=(not a)
INVERTER operation as CSA. select_xor decides what is to be scanned in depending
on the pattern which is to be applied to the combinational logic (which has been provided by the ATPG tool)
Proof of the suitablity of the proposed architecture.
To check whether buffer-insertion or inverter insertion is beneficial for lower scan-in switching activity , counting of patterns for ISCAS'89 circuits was done to find out the number of patterns that give less toggling with buffer insertion and those that give less toggling with inverter insertion.
3 kinds of pattern filling(0 fill,repeat fill,random fill) were tried out.
Procedure applied for random fill
Step 1:
Find the pattern which needs to be scanned-in as per CSA(inverter insertion) from the pattern given by ATPG tool (after random fill)
Step 2:
Count the number of toggles in the pattern found out in Step 1. Step 3:
Count the number of toggles present in the pattern itself. Step 4:
if Step 2 count is less then this pattern needs inverter insertion otherwise this pattern needs buffer insertion(that is to say it is to be scanned-in as it is).Mark this pattern accordingly in the appropriate category.
Repeat the above steps for all the patterns and then do counting of patterns in the two Categories.
ISCAS'89 circuits: Random fill
circuit Total patterns Less-toggles-buffer Less-toggles-inv.
s1196 148 74 74
s9234 130 56 74
s38417 335 212 123
s38584 125 57 68
s15850 115 55 60
s13207 128 63 65
ISCAS'89 circuits:Repeat fill,0 fill
circuit patterns buffer inverter
s13207 128 128 0
s15850 115 113 2
circuit patterns buffer inverter
s13207 128 125 3
s15850 115 112 3
0 fill Repeat fill
DISCUSSION
For random fill,The number of patterns which give less less toggles with buffer and those which give less less toggles with inverter is almost equal suggesting that almost for half of the patterns we can save the unnecessary switching activity arising out of the inverter insertion as propsed in CSA architecture.
Experiment employing TetraMax can give exact fault coverage for different kinds of test sets.
Reference
Kyoung Youn Cho, Subhasish Mitra, and Edward J. McCluskeyCalifornia Scan Architecture for High Quality and Low Power Testing,INTERNATIONAL TEST CONFERENCE(ITC),2007.
THANKS FOR YOUR ATTENTION
Slide 1Slide 2Slide 3Slide 4Slide 5Slide 6Slide 7Slide 8Slide 9Slide 10Slide 11Slide 12Slide 13Slide 14Slide 15Slide 16Slide 17Slide 18Slide 19Slide 20Slide 21Slide 22Slide 23Slide 24Slide 25