ISSN 2394-3777 (Print)

ISSN 2394-3785 (Online)

Available online at www.ijartet.com
International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)
Vol. 2, Issue 10, October 2015

Decimal Matrix Code for Enhanced

Memory Reliability against Multiple
Cell Upsets
T.Dhivya1, M.Mohankumar2, Swaminathan Veerapandian3
PG Student, ECE Dept, Sri Eshwar college of Engineering, Kinathukadavu, Coimbatore1
Assistant professor, ECE Dept, Sri Eshwar college of Engineering, Kinathukadavu, Coimbatore2
Employee, Network executive in Tata communications Ltd., Coimbatore3

Abstract: Transient multiple cell upsets MCUs are becoming major issues in their liability of
memories exposed to radiation environment. To prevent MCUs from causing data corruption,
more complex error correction codes(ECCs) are widely used to protect memory, but the main
problem is that they would require higher delay over head. Moreover, the encoder-reuse
technique (ERT) is proposed to minimize the area over head of extra circuits without disturbing
the whole encoding and decoding processes. The main issue is that they are double error
correction codes and the error correction capabilities are not improved in all cases. In this paper,
novel decimal matrix code (DMC) based on divide-symbol is proposed to enhance memory
reliability with lower delay over head. The proposed DMC is compared to well known codes
such as existing Hamming codes, Punctured different set PDS codes. The only drawback is that,
it requires more redundant bits for memory protection. The obtained results show that the mean
time to failure (MTTF) of the proposed scheme is 452.9%, 154.6% and 122.6% of Hamming,
MC, and PDS, respectively. Recently, matrix codes (MCs) based on Hamming codes have been
proposed for memory protection. The proposed DMC utilizes decimal algorithm to obtain the
maximum error detection capability. ERT uses DMC encoder itself to be part of the decoder. At
the same time, the delay overhead of the proposed scheme is 73.1%, 69.0% and 26.2% of
Hamming, MC and PDS respectively.
Keywords: Decimal algorithm, error correction codes (ECCs), mean time to failure (MTTF),
memory, multiple cells upsets(MCUs).
I.

INTRODUCTION

As CMOS technology scales down to nano scale

and memories are combined with an increasing
number of electronic systems the soft error rate in
memory cells is rapidly increasing, especially when
memories operate in space environments due to
ionizing effects of atmospheric neutron, alpha-

particle and cosmic rays. Although single bit upset

is a major concern about memory reliability,
multiple cell upsets (MCUs) have become a serious
reliability concern in some memory applications. In
order to make memory cells as fault-tolerant as
possible, some error correction codes (ECCs) have
19

All Rights Reserved 2015 IJARTET

ISSN 2394-3777 (Print)

ISSN 2394-3785 (Online)
Available online at www.ijartet.com
International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)
Vol. 2, Issue 10, October 2015

been widely used to protect memories against of

terrors for years. For example, the Bose
Chaudhuri Hocquenghem codes, ReedSolomon
codes and punctured difference set (PDS) codes
have been used to deal with MCUs in memories.
But these codes require more area, power, and delay
overheads since the encoding and decoding circuits
are more complex in these complicated codes. Inter
leaving technique has been used to restrain MCUs,
Which rearrange cells in the physical arrangement
to separate the bits in the same logical word into
different physical words. However, interleaving
technique may not be practically used in contentaddressable memory (CAM), because of the tight
coupling of hardware structures from both cells and
comparison circuit structures.
Built-in current sensors (BICS) are proposed to
assist with single-error correction and double-error
detection codes to provide protection against
MCUs. However, this technique can only correct
two errors in a word. More recently in 2-D matrix
codes (MCs) are proposed to efficiently correct
MCUs per word with a low decoding delay in which
one word is divided into multiple rows and multiple
columns in logical. The bits per row are protected
by Hamming code while parity code is added in
each column. For the MC based on Hamming, when
two errors are detected by Hamming the vertical
syndrome bits are activated so that these two errors
can be corrected.
As a result MC is capable of correcting only two
errors in all cases. In an approach that combines
decimal algorithm with Hamming code has been
conceived to be applied at software level. It uses
addition of integer values to detect and correct soft
errors. The results obtained have shown that this
approach have a lower delay overhead over other
codes. In this paper, novel decimal matrix code
(DMC) based on divide-symbol is proposed to
provide enhanced memory reliability. The proposed
DMC utilizes decimal algorithm (decimal integer
addition and decimal integer subtraction) to detect
errors. The advantage of using decimal algorithm is

that the error detection capability is maximized so

that the reliability of memory is enhanced. Besides
the encoder-reuse technique (ERT) is proposed to
minimize the area overhead of extra circuits
(encoder and decoder) without disturbing the whole
encoding and decoding processes, because ERT
uses DMC encoder itself to be part of the decoder.
This paper is divided into the following sections.
The proposed DMC is introduced and its encoder
and decoder circuits are present in Section II. This
sectional so illustrates the limits of simple binary
error detection and the advantage of decimal error
detection with some examples. The reliability and
over heads analysis of the proposed code are
analyzed in Section III. In Section IV the
implementation of decimal error detection together
with BICS for error correction in CAM is provided.
Finally some conclusions of this paper are discussed
and shared in Section V.

II.P ROPOSEDDMC

In this section, DMC is proposed to assure

reliability in the presence of MCUs with reduced
performance over heads, and a 32-bit word is
encoded and decoded as an example based on the
proposed techniques.
A. Proposed Schematic of Fault-Tolerant
Memory
The proposed
schematic
of fault-tolerant
memory is depicted in Fig.1. First during the
encoding (write) process, information bits are fed to
the DMC encoder and then the horizontal redundant
20

All Rights Reserved 2015 IJARTET

ISSN 2394-3777 (Print)

ISSN 2394-3785 (Online)
Available online at www.ijartet.com
International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)
Vol. 2, Issue 10, October 2015

and m =4 are utilized to construct DMC.

Corrector

B. Proposed DMC Encoder

In the proposed DMC, first, the divide-symbol and
arrange-matrix ideas are performed. i.e., the N-bit

Locator

maximize the correction capability and minimize

the number of redundant bits. For example, in this
case, when k=2*2 and m=8,only one bit error can
be corrected and the number of redundant bits is 40.

Syndrome
(HandS)

bits H and vertical redundant bits V are obtained

from the DMC encoder. When the encoding process
is completed the obtained DMC code word is stored
in the memory. If MCUs occur in the memory.
These errors can be corrected in the decoding (read)
process. Due to the advantage of decimal algorithm,
the proposed DMC has higher fault-tolerant
capability with lower performance overheads. In the
fault-tolerant memory the ERT technique is
proposed to reduce the area overhead of extra
circuits and will be introduced in the following
sections.

word is divided into k symbol so fm bits (N =km),

When k=44 and m=2, 3-bit errors can be corrected

and the number of redundant bits is reduced to 32.
However, when k=24 and m=4, the maximum
correction capability is upto 5 bits and the number
of redundant bits is 36. In this paper, in order to
enhance the reliability of memory, the error
correction capability is first considered, so k =24
The horizontal redundant bits H can be obtained
by decimal integer addition as follows:

and these symbols are arranged in a k1k2 2-D H4H3H2H1H0=D3D2D1D0+D11D10D9D8

(1)
matrix (k=k1k2), where the values of k1 and k2
represent the numbers of the horizontal redundant H9H8H7H6H5=D7D6D5D4+D15D14D13D12
bits H are produced by performing decimal integer (2)
addition of selected symbols per row. Here, each And similarly for the horizontal redundant bits
symbol is regarded as a decimal integer. Third, the H14H13H12H11H10 and H19H18H17H16H15
vertical redundant bits V are obtained by binary where + represents decimal integer addition.
operation among the bits per column. It should be 3.
noted that both divide-symbol and arrange-matrix For the vertical redundant bits V, we have
are implemented in logical instead of in physical.
Therefore, the proposed DMC does not require
V0 =D0XORD16 (3)
changing the physical structure of the memory. To
V1 =D1XORD17 (4)
explain the proposed DMC scheme, we take a 32bit word as an example as shown in Fig. 2. The cells
from D0 to D31 are information bits. This32-bit And similarly for the rest vertical redundant bits.
The encoding can be performed by decimal and
word has been divided into eight symbols of 4-bit.
binary addition operations from (1) to (4). The
k1=2 and k2=4 have been chosen simultaneously. encoder that computes the redundant bits using
H0 H19 are horizontal check bits. V0 through V15 multibit adders and XOR gates is shown in Fig. 3. In
are vertical check bits. However, it should be this figure, H19H0 are horizontal redundant bits,
mentioned that the maximum correction capability
(i.e., the maximum size of MCUs can be corrected) V15V0 are vertical redundant bits, and the
and the number of redundant bits are different when remaining bits U31U0 are the information bits
the different values for k and m are chosen. which are directly copied from D
31 to D0. The
Therefore, k and m should be carefully adjusted to
21
All Rights Reserved 2015 IJARTET

ISSN 2394-3777 (Print)

ISSN 2394-3785 (Online)
Available online at www.ijartet.com
International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)
Vol. 2, Issue 10, October 2015

enable signal En will be explained in the next

section.
II.
Proposed DMC Decoder
To obtain a word being corrected, the decoding
process is required. For example, first, the received
redundant information bits D. Second, the

the stored codeword has original information bits in

symbol 0 where no errors occur. When
H4H3H2H1H0 and S3S0 are non zero, the induced
errors (the number of errors is 4 in this case) are
detected and located insymbol0, and then these errors

can be corrected by D0correct =D0S0 (7) The

proposed DMC decoder is depicted in Fig.4, which is
horizontal syndrome bits
made up of the following sub-modules, and each
H4H3H2H1H0 and the vertical syndrome bits
executes a specific task in the decoding process:
S3S0 can be calculated as follows:
Syndrome calculator, error corrector. It can be
observed from this figure that the redundant bits

0
information
H4H3H2H1H0=H4H3H2H1H H4H3H2H1H0 must be recomputed from the received

(5)
the original set of redundant
bits D and compared to

(6)
S0 =V0V0
bits in order to obtain the syndrome bits H and S. The
And similarly for the rest vertical syndrome bits, error locator uses H and S to detect and locate which
bits some error occur in. Finally the error can be
whererepresents decimal integer subtraction.
corrected by inverting values of error bits.
When H4H3H2H1H0 and S3S0 are equal to zero,

D. Limits of simple binary detection:

For the proposed binary error detection
technique in [13], although it requires low
redundant bits, its error detection capability is
22
All Rights Reserved 2015 IJARTET

ISSN 2394-3777 (Print)

ISSN 2394-3785 (Online)
Available online at www.ijartet.com
International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)
Vol. 2, Issue 10, October 2015

limited. The main reason for this is that its error

detection mechanism is based on binary.

Then horizontal syndrome bits are

H4 H3 H2 H1 H0=H4 H3 H2 H1 H0-H4 H3
H2 H1 H0 =10110-10010=00100.
The decimal value of H4H3H2H1H0 is not 0,
which represents that errors are detected and
located in symbol 0 or symbol 2. Subsequently,
the precise location of the bits that were flipped
can be located by using the vertical syndrome bits
S3S0 and S11 S8. Finally, all these errors can
be corrected by (7). Therefore, based on decimal
algorithm, the proposed technique has higher
tolerance capability for protecting memory
against MCUs.

We illustrate the limits of this simple binary

error detection [13] using a simple example. Let
us suppose that the bits B3 , B2, B1, and B0 are
original information bits and the bits C0 and C1
are redundant bits shown in Fig.5. The bits C0
and C1 are obtained using the binary algorithm.
E. Advantage of Decimal Error Detection
In the previous discussion, it has been shown
that error detection [13] based on binary
algorithm can only detect a finite number of
errors. However, when the decimal algorithm is
used to detect errors, these errors can be detected
so that the decoding error can be avoided. The
reason is that the operation mechanism of
decimal algorithm is different from that of binary.
The detection procedure of decimal error
detection using the proposed structure shown in
Fig.2 is fully described in Fig.6. First of all, the
horizontal redundant bits H4H3H2H1H0 are
obtained from the original information bits in
symbols 0 and 2 according to (1)
H4 H3 H2 H1 H0=D3 D2 D1 D0+D11 D10
D9 D8=1100+0110=10010
H4 H3 H2 H1 H0=D3 D2 D1 D0+D11 D10 D9
D8=0111+1111=10110

The proposed DMC can easily correct upset

softype1, 2, and 3, because these are the essential
property of DMC: any types of single-error and
multiple-error corrections in two consecutive
symbols. Upsets of types 4 and 5 introduced in
Fig.7 are also corrected because the multiple
errors per row can be detected by the horizontal
syndrome bits (see Fig. 6). These show that the
proposed technique is an attractive option to
protect memories from large MCUs. However,
for the upsets of type4 and 5, It is important to
recognize that it can result in decoding error
when the following pre requisite factors are
achieved simultaneously (this error is typical of
its kind).
1) The decimal integer sum of information bits
in symbols 0 and 2 is equal to 2m1.
23

All Rights Reserved 2015 IJARTET

ISSN 2394-3777 (Print)

ISSN 2394-3785 (Online)
Available online at www.ijartet.com
International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)
Vol. 2, Issue 10, October 2015

2) All the bits in symbols 0 and 2 are upset.

0
The more detailed
explanation is shown in
Fig.8. Assuming that these two factors have been
achieved, according to the encoding and
decoding processes of DMC, H4H3H2H1H0 , and

H4H3H2H1H are computed, as follows:

H4H3H2H1H0=D3D2D1D0+D11D10D9D8
=0110+1001
=01111

(17)

H4H3H2H1H =D3D2D1D0 +D11D10D9D

8

(18)

Then
the
horizontal
syndrome
H4H3H2H1H0 can be obtained

RELIABILITY AND OVERHEADS

ANALYSIS
In this section, the proposed DMC has been
implemented in HDL, simulated with Model Sim
and tested for functionality by given various
inputs.
The area, power, and critical path delay of extra
circuits have been obtained. For fair
comparisions, Hamming, PDS [9], and MC [15]
are used for references. Here, the usage of (64,
45) PDS is a triple-error correction code [9] and
its information bits is shorted to 32 from 45 bits.
Fault Injection:
The correction coverage of PDS [9], MC [15],
Hamming, and the proposed DMC codes are
obtained from one million experiments.

=1001+0110
=01111.

III.

bits

H4H3H2H1H0=H4H3H2H1H H04H3H2H1H0
=0111101111
=00000
This result means that no errors occur in
symbols 0 and 2 and memory will suffer a failure.
However, this case is rare.

These results show how our proposed technique

provides single and double-error correction, but
can also provide effective tolerance capabilities
against large MCUs that exceed the performance
of other codes.

24
All Rights Reserved 2015 IJARTET

ISSN 2394-3777 (Print)

ISSN 2394-3785 (Online)
Available online at www.ijartet.com
International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)
Vol. 2, Issue 10, October 2015

Reliability Estimation:
The reliability of our proposed code can be
analyzed in terms of the mean time to failure
(MTTF). It is assumed that MCUs arrive at
memories following a Poisson distribution.
MTTF is given by MTTF=J(t). dt
Table II shows MTTFs of different codes for
different event arrival rate . In this table, we can
see that the proposed scheme has higher MTTF
bymore than 122.6%, 154.6%, and 452.9%
compared to PDS [9], MC [15], and Hamming,
respectively

25
All Rights Reserved 2015 IJARTET

ISSN 2394-3777 (Print)

ISSN 2394-3785 (Online)
Available online at www.ijartet.com
International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)
Vol. 2, Issue 10, October 2015

In general cases, for proposed technique

it can be inferred that the larger the word
widths, the higher the tolerance capabilities and
better the reliabilities.
C.

Overheads Analysis:

The area and power overheads of PDS are

1170.9% and 2047.2% of the proposed
scheme, respectively. The delay overhead of
DMC is 26.2%, 69.0%, and 73.1% of PDS
[9], MC [15], and Hamming respectively. This
indicates that the memory with the proposed
scheme performs faster than other codes.
Different decoding algorithms could result in
different overheads.
The decoding algorithm of PDS [9] is
more complex than that of other codes. Thus,
it has maximum area, power, and delay
overheads. However, for the proposed DMC, its
decoding algorithm is simple with minimal
overhead. The issue is that the proposed
technique requires more redundant bits
compared with other codes.
IV. CONCLUSION
In this paper, novel per-word DMC was
proposed to assure the reliability of memory.
The proposed protection code utilized decimal
algorithm to detect errors, so that more errors
were detected and corrected. The obtained
results showed that the proposed scheme has a
superior protection level against large MCUs
in memory. Besides, the proposed decimal
error detection technique is an attractive
opinion to detect MCUs in CAM because it
can be combined with BICS to provide an
adequate level of immunity. The only
drawback of the proposed DMC is that more
redundant bits are required to maintain higher
reliability of memory, so that a reasonable
combination of k and m should be chosen to

maximize memory reliability and minimize

the number of redundant bits based on
radiation
experiments
in
actual
implementation. Therefore, future work will
be conducted for the reduction of the
redundant bits and the maintenance of the
reliability of the proposed technique
REFERENCES
[1]

[2]

[3]

[4]

[5]

[6]

[7]

D. Radaelli, H. Puchner, S. Wong, and S.

Daniel, Investigation of multi-bit up sets in
a 150nm technology SRAMdevice,IEEE
Trans. Nucl. Sci., vol. 52, no. 6 ,pp. 2433
2437,Dec.2005.
E.Ibe, H.Taniguchi, Y.Yahagi, K. Shimbo,
and T.Toba, Impact of scaling on neutron
induced soft error in SRAM sfroman 250nm
to
a
22nm
design
rule,
IEEETrans.ElectronDevices, vol.57, no.7,
pp.15271538,Jul.2010.
C. Argyridesand D. K. Pradhan, Improved
decoding algorithm for high reliable reed
muller coding,in Proc .IEEE Int. Syst. On
Chip Conf., Sep.2007, pp. 9598.
A. Sanchez-Macian, P. Reviriego, and J. A.
Maestro, Hamming SEC-DAED and
extended hamming SEC-DED-TAED codes
through selective shortening and bit
placement, IEEE Trans. Device Mater.
Rel.,to be published.
S.Liu, P.Reviriego, and J.A.Maestro,
Efficient majority logic fault detection
with difference- set codes for memory
applications, IEEE Trans. Very Large
ScaleIntegr
.(VLSI)
Syst.,vol.20,no.1,pp.148156, Jan.2012.
M.Zhu, L.Y.Xiao, L.L.Song, Y.J.Zhang,
and H.W.Luo, New mix codes for multiple
bit up sets mitigation in fault-secure
memories, Microelectron.J.,vol. 42, no.3
,pp. 553561,Mar.2011.
R.Naseer and J.Draper, Parallel double
error correcting code design to mitigate
multi-bit
up
sets
in
SRAMs,
26

All Rights Reserved 2015 IJARTET

ISSN 2394-3777 (Print)

ISSN 2394-3785 (Online)
Available online at www.ijartet.com
International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)
Vol. 2, Issue 10, October 2015

[8]

[9]

[10]

[11]

[12]

[13]

[14]

inProc.34thEur.Solid-State Circuits, Sep.

2008, pp. 222225.
G.Neuberger,D.L.Kastensmidt,andR.Reis,
An automatic technique For optimizing
Reed-Solomon codes to improve fault
tolerance
in
memories,IEEE
DesignTestComput.,vol.22,no.1,pp.5058,
Jan.Feb.2005.
P.Reviriego, M.Flanagan, and J.A.Maestro,
A (64, 45) triple error correction code for
memory applications, IEEE Trans. Device
Mater. Rel.,vol.12 ,no.1, pp.101106,
Mar.2012.
S.Baeg, S.Wen, and R.Wong, Inter
leaving distance selection with a soft
error failure model, IEEE Trans. Nucl
.Sci., vol. 56, no. 4, pp.2111
2118,Aug.2009.
K.Pagiamtzisand
A.Sheikholeslami,
Content address able memory (CAM)
circuits and architectures: A tutorial and
survey, IEEE J. Solid-State Circuits,
vol.41, no.3, pp.712727,Mar.2003.
S.Baeg, S. Wen, and R. Wong, Minimizing
soft errors in TCAM devices: A
probabilistic approach to determining
IEEE
scrubbing
intervals,
Trans.CircuitsSyst.I
,Reg.
Papers,vol.57,no.4,pp.814822, Apr.2010.
P.Reviriego and J. A. Maestro, Efficient
error detection codes for multiple-bit upset
correction in SRAMs with BICS, ACM
Trans
.Design
Autom.Electron.Syst.,vol.14,no.1,pp.18:1
18:10,Jan.2009.
C.Argyrides, R.Chipana, F.Vargas, and
D.K.Pradhan, Reliability analysis of H-tree
random access memories implemented with
built in current sensors and parity codes for
multiple bitup set correction, IEEE
Trans.Rel.,vol.60,no.3,pp.528
537,Sep.2011.

Authors Profile
Miss.
T.
DHIVYA
pursuing Master degree in
VLSI Design from Sri
Eshwar
College
of
Engineering, Coimbatore.
Area of interests are Low
Power VLSI and Testing
of VLSI.
Mr. M. MOHANKUMAR
is Assistant professor at Sri
Eshwar
College
of
Engineering,
Coimbatore.
His Specialization area is
VLSI
Design.

Mr. SWAMINATHAN
VEERAPANDIAN
is
working
as
network
executive
in
Tata
Communications Ltd. His
specialization area is
Networking Systems.

27
All Rights Reserved 2015 IJARTET