Improvement in Error Resilience in BIST using hamming code

82 International Journal for Modern Trends in Science and Technology
International Journal for Modern Trends in Science and Technology
Volume: 02, Issue No: 10, October 2016
http://www.ijmtst.com
ISSN: 2455-3778
Improvement in Error Resilience in BIST using
hamming code
Radharapu Ravivarma1
| M. Sanjay 2
1PG Scholar, Vaagdevi College of Engineering, Telangana.
2Assistant Professor, Vaagdevi College of Engineering, Telangana.
To Cite this Article
Radharapu Ravivarma and M.Sanjay, Improvement in Error Resilience in BIST using hamming code , International
Journal for Modern Trends in Science and Technology, Vol. 02, Issue 10, 2016, pp. 82-87.
In the current scenario of IP core based SoC, to test the CUT we need to communication link between
Circuit Under Test and ATPG, so before applying to actual DUT. If there is a problem with this link, there may
be a lip in bit of test data. Compared to original test data, if there is a bit lip in the original data, the codeword
may change and hence the decompressed data will have a large number of bit deviation. This deviation in
bits can severely degrade the test quality and overall fault coverage which may affect yield. The error
resilience is the capability of the test data to resist against such bit lips. Here in this paper, the earlier
methods of error resilience is compared and a Hamming code based error resilience technique is proposed to
improve the error resilience capacity of compressed test data. This method is applied on Huffman code based
compressed test data of widely used ISCAS benchmark circuits. The fault coverage measurement results
show the effectiveness of the proposed method. The basic goal here is to survey the effect of bit lips on fault
coverage and prepare a platform for further development in this avenue.
KEYWORDS: Automatic Test Equipment (ATE), bit-lip; Compression; Fault Tolerance; Hamming code; Fault
Coverage; Area Overhead;, Bits Overhead
Copyright © 2016 International Journal for Modern Trends in Science and Technology
All rights reserved.
I. INTRODUCTION
Considering the today's scenario of fast
time-to-market and increasing test cost compared
to manufacturing cost, the use of Intellectual
Property (IP) course for gathering complex
functionality has been common. [1]. IP core based
system on chip(SoC) is becoming new paradigm in
the electronics based industry for its reusability
and ability to execute rich set of functionality in a
very short time. Multiple core on a SoC adds
complexity to the design, so it's become challenging
to test each core, as all core have different 1I0S,
different test pattens and different scan chain
lengths, etc. In general, SoC test challenge includes
the dificulty of distributing design inside the SoC
and its test development, the difficulty of accessing
the test in embedded core and finally optimization
of the SoC level test.
A test set for a SoC test includes test vectors for
each core. The total volume of these data can be
very high, can be in the scale of a gigabits which
need high storage facility, but on another way, it
increases the entire cost of the test. To highlight
these problems, test data compression techniques
are introduced. [2] [3].
Recently, many test data compression
techniques are developed to reduce test data
volume. For example, statistical test data
compression techniques like run length code,
frequency directed run length code [4][5][6],
Golomb code[5][7] Extended FDR [5], Modified
ABSTRACT

Radharapu Ravivarma and M.Sanjay : Improvement in Error Resilience in BIST using Hamming Code
Extended EFDR [5], VIHC coding[8] etc. The
comparison of these techniques is presented in [9
HDR]. Among such various test data compression
method like broadcast scan based methods, linear
decompression based methods and the code based
methods, the code based compression method inds
the most suitable for IP cores where the internal
architecture of core is hidden rom system
Integrator. [9].
When the compressed data is being transferred
rom ATE to SoC, if there is any distortion or noise
in the communication link, it may happen that the
test data bit may lip. This bit lip even though may
not be so serious for uncompressed data, but gives
serious results in case of compressed data because
it may change one or many codewords in
compressed data and correspondingly
regenerated-decompressed data will be different
from original data. When such deviated data will be
applied to DUT for testing, definitely will not give
the test quality which was intended. The fault
coverage may be affected and hence the yield will
be also affected.
The capability of test data to resist the effects of
such bit lips is called error resilience. Fault
endurance, reliability and security are the major
concern for resilience of test data. So the error
resilience can be defined as the ability of a design
under test (DUT) to execute proper unction in the
presence of bit lip error, bit deletion, burst error
and missing fragmentations and to recover from
any service degradation. These stuffs are essential
in a testing where the validity of data depends on
the reliable, secure and correct function of the
device under test. Considered systems include, but
are not limited to, infrastructures, computer
networks, including adhoc and mesh networks,
web-based systems, or service- oriented
architectures, embedded systems, manufacturing
systems, control systems and more. Much work
has been done already on evaluation, analysis and
enhancement of error resilience of test data, but a
lot remains to be done. The combination of
performance and dependability, performability,
has been widely studied. But aspect, like,
improvement in error resilience is still ongoing
work. As the systems are changing day by day as
per its characteristics, it's a high need to expand
work in this area device under test. Considered
systems include, but are not limited to,
inrastructures, computer networks, including
adhoc and mesh networks, web-based systems, or
service- oriented architectures, embedded systems,
manufacturing systems, control systems and
more. Much work has been done already on
evaluation, analysis and enhancement of error
resilience of test data, but a lot remains to be done.
The combination of performance and
dependability, performability, has been widely
studied. But aspect, like, improvement in error
resilience is still ongoing work. As the systems are
changing day by day as per its characteristics, it's a
high need to expand work in this area The paper
has addressed methods and tools to describe,
measure, evaluate benchmark and improve
resilience concening resilience of test data. Here in
this paper, we have calculated the effect of bit lips
on fault coverage of widely cited ISCAS benchmark
circuits for both the case : I. when normal
(uncompressed) data is used through serial link
and 2. When the compressed data is used and the
comparisons are analyzed. Further we have
proposed the Hamming code base approach to
improve the error resilience and its effectivity is
analyzed by calculating the bit overhead and
improvement in fault coverage. The basic goal here
is to prepare a platform for further development to
improve error resilience capability in case of
compressed test data being used for IP core based
SoC
II. THE SIGNIFICANCE OF BIT FLIP N TEST
DATA
A. Why Bit Flp Occurs?
Bit lips can occur in many ways during the
transmission of test data from ATE to OUT. Noise
can affect the interfacing line between the ATE and
design under test, which can raise the errors
because of bit lip. Moreover, Bit lips can also be
generated when test inputs are generated as a pin
waveform. So it can be said that bit lip can be
reduced if the ATE test program and the DUT is
fully debugged. However, as testing is done at very
high speed, eror in few bits(bit lips) must be
considered as unavoidable in ATE environments
[10].
B. fect of Bit Flip on Fault Coverage
During the development of the test program,
Bit-lips can produce diverse effects over its entire
testing environment. Throughout the
manufacturing test, a bit-lip affects the test data in
such a way that the coverage of inaccurate test set
can be reduced (as affected by bit-lips). As a result,
there is a hike in cost incorporated with debugging
and development, while decreases productivity.

C. Experimental Set-up to measure the efect ofBit-lip on
Fault Coverage
Here the full scan version of the ISCAS'89 circuits
is used as a platform to evaluate different bit lip
effects and associated fundamentals. The
experiments were performed on test data generated
by MINTEST tools. In the given test cubes, all
unspecified bits are filled with MT ill technique and
then re-ordered using the Artificial Intelligence (AI)
based algorithm[II][12]. Generally the occurrence of
bit lips is considered in two ways i.) A bit lip
probability ii.) Bit lip count. In this paper, we are
focusing mainly on bit lip count. The bit lip
locations are uniformly spread over the entire test
vector. So the bit lips are uniformly distributed to
the entire bit stream and can be evaluated in all
conditions. Here, for simplicity bit lip count of 1,2 5
and 10 are used for experimental framework. Here
each bit lip count is repeated for 25 times and
average fault coverage is calculated. The Huffman
compression technique is used for data
compression with symbol size of 4 bits [13].
D. fect of Bit Flip on Uncompressed Data
As we know, bits in compressed test data carries
more information as compared to original bit
stream, So bit lip is important for test data
compression circuits. To understand the minimum
effect of bit lip, it is understand that fault coverage
loss with original test data due to bit lip represents
lower bound on the eror resilience. For the proper
understanding, bit lips are uniformly injected in
the test data and corresponding fault coverage is
measured. Table I shows the fault coverage due to
bit lip in uncompressed test data. The average
coverage loss is 0.1 to 0.32% for 1 bit, 2 bit,S bit
and 10 bit lips.
III. EFFECT OF BIT FLIP ON COMPRESSED
TEST DATA
Assume that there are V test vectors which are
the source of combinational and sequential test
data. In the compression technique discussed
above, test vectors are partitioned into given
symbol size ((sill ) S I S S) where S is the total
number of different symbols. Ouring compression,
each symbol Sj is mapped to its associated
codeword. The compressed test data can be written
as an ordering n of the codeword (C,ll),
Cn(2),....C,lN)), where N is the total number of
symbols per codeword in the test set. The different
effects with bit lips are described below.
A. Propagation and sht efect
As it is discussed, bit lip can change the
compressed sequence in a complicated manner,
which depends on a number of symbols,
compression ratio, length of symbol or codeword
and ordering technique used. In general, bit lip can
affect codeword in two ways.
Codeword which is affected is partitioned into(K �
1) valid codeword.
Codeword which is affected is partitioned into(K �
0) valid codeword and so called dangling suix,
which is not a valid code word.
Example: Assume thatcodeword for the Huffman
compression is given as 0,10,110,I11.
Now if there is a bit lip in 15t codeword as given
in example, then it becomes'l' which is not a valid
codeword. If there is a bit lip in the 1st bit of 2nd
codeword then it becomes "00", Which gives two
different codewords '0'-'0'. But if there is a bit lip in
2nd bit of 2nd codeword then it becomes "11"
which is not valid codeword (only dangling sufix).
As shown in case I, if there is a bit lip in the
codeword Cn(i), then codeword C[(i+I) to Cn(N) will
remain unaffected. So the bit lip will generate K
additional codeword C,li-I) and Cn(i+I). As a result,
correct sequence of codeword will start rom C,li+1),
shited by K additional codeword as shown in fig 3.
Fig 3. Conceptual view of the shit effect due to a bit lip
Now, as shown in case II, if there is a bit lip in the
codeword C,li) and if bit lip creates a dangling suix,
then that dangling sufix will form a valid codeword
with bits of remaining codeword starting from
Cn(i+I). This phenomena is called as propagation of
bit lipto codeword C,li+1). Now, if codeword C,li+1)
also let with dangling sufix, then that dangling suix
will make codeword with bits of C,li+2) codeword. If
Cn(i+j) doesn't have dangling sufix, then codewords
C[(i+j+1) to C[(N) will be unaffected which is shown
in fig 4.
S820 99. 14 98. 92 98. 92 98. 92
S832 97. 17 97. 17 97. 17 97. 17
S838 99. 67 99. 67 99. 67 99. 67

Above two cases signifies the relationship
between shit and propagation effect due to bit lip.
In the 1st case, codeword Cn(i) is lost and forms
another valid codeword which shits remaining code
words by at least one codeword.
In the 2nd case, due to bit lip, codeword j+I are
lost and dangling suffix will make a codeword by
using bits of next codeword. So the remaining code
words will shift by minimum one codeword. So it
can be said that propagation and shit effect can be
differentiated in terms of the number of lost
codeword. As it can be seen in the diagram, shit is
the special case of propagation where only one
codeword is lost, propagation always comes with
shit where number of lost codeword depends on the
location of the bit lip.
B. Synchronization loss
After decompression, corrupted sequence of bits
must appear in the same sequence as it would
there in case of no bit flipping. In case of bit
flipping, decompressed sequence will be disturbed
by the corrupt sequence of bits. If these corrupt bit
sequences are different than original one, then
there is a loss of synchronization and hence
decoding of the received code will be difficult.
C. Bit Flip in case Differentiated Vectors
In the compression technique described above,
data is first of all, re-ordered and then
differentiation is done to maximize compression.
Assume that 01, O2, 03 • . • . • . . • . . On is the
difference vector which can be defined as
01=V1
O2= V2 XOR VI
O"=V,, XOR V,,_1
At the decompression side assigned vectors are
retrieved in the following way.
VI = DI
Vz = Dz XOR VI
V" = 0" XOR 0,,-1 XOR ............02 XOR 01
Now in this process, if bit lip occurs in bit OJ, then
it will affect on all subsequent vectors during
decompression and all vector rom i to n will be
affected.
D. Effect of Bit Flip on Fault Coverage of
Compressed Data
Assume that the test data is compressed with the
compression ratio of 30%. That means, there is
30% minimization in data volume. So it can be said
that each 70 bits in the compressed test set can be
expanded into 100 bits of the initial uncompressed
test set. These bits will be scattered over entire test
data, which can affect the multiple test vectors and
degrade the fault coverage. Table II shows the
average fault coverage with bit lip in Huffman
compressed ISCAS circuit test data.
IV. PROPOSED METHOD TOIMPROVEERROR
RESILENCE N CASE OF BIT FLIP
In the previous section, importance of bit lip and
its negative impact on fault coverage was
described. As a solution of that, the new technique
is developed to improve the fault coverage and
hence error resilience with compressed test data. A
parity bit based error resilience method is
described in [14]. Here, in this paper, the advance
method for error resilience is proposed based on
Hamming distance.
A. Hamming Code based Technique
In communication, The Hamming code is called
as linear error code which detects and corrects
errors. At the most, it can detect two bits of error
and corrects one bit. If the hamming distance
between received and transmitted bit pattern is

less than one, then it is called as reliable
communication. Hamming code is like a binary
linear code. For each code m : 2, there will be m
parity bits and 2m -m+1 message bits. Hamming
code is the best example of the perfect code which
means this code matches the theoretical upper
bound on different code words for the given bits
and its strength to correct the error bits. If more
error correcting bits are added along with the
message and if such bits are arranged to get
different error results in different incorrect bits,
then bad bits can be identified. For example, in (7,
3) Hamming code, there is a chance of error in 7
single bits. So in that 3 parity bits not only signifies
the error in data, but also inds the position of the
error.
B. General algorithm
1. Signiies the bit numbers in binary form. I.e.
for 3 bit code, Bit I = 00I, Bit 2 = 01
2. Bit positions which are powers of 2 are
considered as parity bits. e.g. 2° - 1st bit as
parity bit
3. Consider all remaining bits as data bits.
4. Parity bits are calculated rom the different
combination of data bits as per the binary
structure of the bit position.
Parity bit I includes the data bits which
have least signiicant bit 'I' in their bit
location. i.e bit 1,3,5,7....
Parity bit 2 includes the data bits which
have least signiicant bit '1' in their bit
location. i.e bit 2,3,6,7,....
5. Calculate parity rom the message bits. For
simplicity, (7,3) hamming code is taken as
an example, where,out of 7 bits, 4 bits are
message bits (0], O2, 03, 04) and 3 bits are
parity bits (C], C2, C3). Assume that
message sequence given here is "0111"
CI = Dlxor D2xor 04 C1 = 0 xor 1 xor 1
C1 = 0
C2 = 01Xor 03xor 04 C2 = 0 xor 1 xor 1
C2 = 0
C3 = 02Xor 03xor 04 C3 = I xor 1 xor I
C3 = I
6. After calculating the parity, again form a
full sequence of hamming code. In our
example it becomes "0001111".
7. Now suppose at the receiver side instead of
"0001111", "0011111" is received. That
indicates the bit flipping at some location
8. At the receiver side, decode the code in
following mannerAl = Clxor 0lxor 02xor 04
AI = 0 xor I xor 1 xor I
Al = 1
A2 = C2xor 01Xor D3Xor D4 A2 = 0 xor 1 xor 1 xor
1
A2 = 1
A3 = C3xor D2xor D3Xor D4 A3 = 1 xor 1 xor 1 xor
1
A3 = 0
Where A), A2 and A3 bits are answer bits.
9. If A], A2 and A3 are zero, then there is no error in
the message. But if any bit is '1' then there is an
error.
10. Arranage A], A2 and A3 bits n reverse order.
A3A2Al = 011. That indicates there is an error at
the 3rd position in fmal hamming sequence. So
now lip that bit and solve the error.
Now apply this algorithm to compress test data.
Divide whole data set in to blocks of 4 bits. For
each 4 message bits calculate the parity bits and
detect and correct error.
V. EXPERIMENTAL RESULTS
So the main advantage of Hamming code is that
it removes bit lip error, so no loss of fault coverage
as shown in table III and hence it improves error
resilience. But the biggest disadvantage of this
method is bits overhead as shown in table IV
Table IV BITS Overhead with Hamming Technique
Bench
mark
Circuit
Bits Human
Communcation
Total Bits
Hamming
Code
Bits
Overhead
S27 18 35 17
S298 406 714
S344 390 680
S349 420 735
S382 714 1253
S386 300 525
S400 798 1400
S420 928 1624

S444 735 1288
S510 402 707
VI. CONCLUSION
Through extensive simulation on the different
benchmark circuit, it is shown that the bit lip in
compressed test data impacts a lot on fault
coverage as compared to uncompressed test data.
But with the proposed Hamming code based
technique, impact of bit flipping on compressed
test data can be significantly nullify. So the
proposed technique can improve the error
resilience but at the cost of more bits overhead.
REFERENCES
[1] Y. Zorian, "Test requirements for embedded
core-based systems and IEEE PI50 0 ," in Proc.IEEE
Int. Test Coj, 1997, pp. 191-199.
[2] Touba NA,"Survey of Test Vector Compression
Techniques", IEEE D esign and Test of Computers,
20 06, 3(4), pp. 294- 30 3
[3] D. Ha M. Ishida and T. Yamaguchi. Compact: A
hybrid method for compressmg test data. IEEE VLSI
Test Symposium, pp. 62-69, 1998.
[4] U Mehta, N D evashrayee, K S D as Gupta,
"Combining Unspeciied Test D ata Bit Fil l ing
Methods and Run Length Based Codes to Estimate
Compression, Pow er nd Area overhead", IEEE
Annual Symposium on VLSI,pp-448,449,20 10
[5] U Mehta, N D evashrayee, K S D as Gupta,
"Run-Length-Based Test Data Compression
Techniques: How Far from Entropy and Power
Bounds?-A Survey", VLSI D esign, Hindawi Publ
ication Corporation, PP-9,20 10
[6] A. Chandra nd K. Chkrabarty. "Frequency-D irected
Run-Length (FD R) codes with appl ication to
system-on-a-chip test data compression," in
Proc.IEEE VLSI Test Symp., 20 0 1, pp. 42-47.
[7] A. Chandra and K. Chakrabarty, "System-on-a-chip
test-data compression nd decompression
architectures based on Gol omb codes," IEEE Trans.
Comput.-Aided Des. Integr. Circuits Syst., vol . 20 ,
no. 3, pp. 355-368, Mar. 20 01.
[8] P. Gonciari, B. AI-Hashimi, and N. Nicolici, "Variabl
e-l ength input Huffman coding for system-on-a-chip
test," IEEE Trans.Comput.-Aided Des.Inter.Circuits
Syst., vol . 22, no. 6, pp. 783-796, Jun. 20 03.
[9] U Mehta, N D evashrayee, K S D as Gupta," Weighted
Transition Based Reordering, Columnw ise Bit Fil l
ing, and D ifference Vector: A Power- Aw re Test D
ata Compression Method", VLSI D esign, Hindaw
i Publ ication Corporation, PP-9,20 11.
[10]B. West, private communication, Credence Corp
[11]H Parmar, U Mehta, " A statistical test data
compression technique with adaptive bit ll ing nd AI
based reordering: optimization for compression and
scan power", Intenational Jounal of VLSI and Sinal
Processing Applications, Vol.I, Issue 2, May2011,pp
-15-24
[12]H Parmar, U Mehta, K D asgupta, N D evashrayee,"
Test D ata Compression Technique for IP Core
based SoC using Artiicial Intel l igence",
ASDAT,Jan 20 11/
[13]M Sharma," Compression using Hufnan Coding",
IJCSNS Intenational Jounal of Computer Science
and Network Securiy, VOL. 10 No. 5, May 2010 .
[14]U Mehta, R Trivedi, N Thakkar, "Error Resil ience in
Case of Test D ata Compression Techniques for IP
Core Based SoC", Intenational Journal of Advance
Research in Engineering and Technology, Vol ume 5,
Issue I, January (2014), page:24-35.

Improvement in Error Resilience in BIST using hamming code

More Related Content

What's hot (14)

Viewers also liked (20)

Similar to Improvement in Error Resilience in BIST using hamming code (20)

Recently uploaded (20)

Improvement in Error Resilience in BIST using hamming code