![TELKOMNIKA, Vol.16, No.2, April 2018, pp. 573~579
ISSN: 1693-6930, accredited A by DIKTI, Decree No: 58/DIKTI/Kep/2013
DOI: 10.12928/TELKOMNIKA.v16i2.6812 573
Received November 29, 2017; Revised February 18, 2018; Accepted February 28, 2018
Detection of Malicious Circuitry Using Transition
Probability Based Node Reduction Technique
Nirmala Devi M*, Irene Susan Jacob, Sree Ranjani R, Jayakumar M
Department of Electronics and Communication Engineering, Amrita School of Engineering, Coimbatore,
Amrita Vishwa Vidyapeetham, Amrita University, India
*Corresponding author, e-mail: m_nirmala@cb.amrita.edu
Abstract
In recent years, serious concerns have been raised against the tampering of integrated circuits
due to outsourcing of circuits for fabrication. It has led to the addition of malicious circuitry known as
Hardware Trojan. In this paper, a transition probability based node reduction technique for faster and
efficient Hardware Trojan (HT) detection has been attempted. In the proposed method, the fact that the
least controllable and observable nodes or the nodes with least transition probability are more vulnerable
as Trojan sites is taken into consideration. The nodes that have lesser activity than the threshold are the
candidate nodes. At each candidate node, segmentation is done for further leakage power analysis to
detect the presence of Trojans. Experimental results observed on ISCAS’85 and ISCAS’89 benchmark
circuits illustrate that the proposed work can achieve remarkable node reduction upto 78.81% and time
reduction upto 58.7%. It was also observed that the circuit activity can be increased by varying the input
probability. Hence, for further reduction in the Trojan activation time, the weighted input probability was
obtained.
Keywords: Hardware security, Hardware trojans, Transition probability, Weighted input probability,
Segmentation
Copyright © 2018 Universitas Ahmad Dahlan. All rights reserved.
1. Introduction
The increase in the complexity of the modern electronics has led to an increase in the
involvement of various firms from around the globe in each phase of the electronics supply
chain. Outsourcing has increased the integrated circuit vulnerability to various malicious
activities and alterations. These modifications termed as Hardware Trojan(HT) has raised
genuine concerns with respect to conceivable dangers to military frameworks, budgetary
foundations, transportation security, and healthcare appliances. HTs when introduced in a
system, may affect the life span of the system, change the functionality or system specification,
leak confidential information or led to denial of service [1-4].
Detection of such malicious alterations is extremely difficult, due to many reasons. To
start with, given the vast number of soft, firm, and hard IP centers utilized as a part of SoCs, and
in addition the high multifaceted nature of today's IP blocks, distinguishing a little malignant
change is extremely difficult. Second, Trojan circuits, by configuration, are normally initiated
under certain conditions, which make them inactive and hard to be recognized utilizing random
or functional stimuli. Third, tests used to identify fabricating shortcomings, for example, stuck -at
and delay faults can not ensure detection of Trojans [5]. Considering these facts, detection of
Trojans utilizing basic investigation would be ineffective.
An efficient method to accelerate the HT detection is the need of the hour and the same
has been proposed in this work. The method is based on the generation of weighted input
patterns to facilitate the activation of the HTs. Moreover, transition probability analysis node
reduction has been attempted in the circuit for faster detection of HTs the fact that the least
controllable and observable nodes or the nodes with least transition probability are perfect
trigger nodes for adding Trojans. Hence by considering only these nodes, the Trojan detection
process can be accelerated with the same efficiency as the case wherein all the nodes are
under observation. For weighted input pattern generation, an iterative procedure is followed
such that the transition probability at each node should be equal to or greater than the user-
defined threshold. The main contributions of this paper can be listed as follows:](https://image.slidesharecdn.com/136812-200821075910/85/Detection-of-Malicious-Circuitry-Using-Transition-Probability-Based-Node-Reduction-Technique-1-320.jpg)
![ ISSN: 1693-6930
TELKOMNIKA Vol. 16, No. 2, April 2018 : 573 – 579
574
a. Generation of weighted input probability set, which reduces Trojan activation time.
b. Efficient and high speed Trojan detection technique by reducing the nodes of
observation.
c. Validation of the effectiveness of Transition probability based Node reduction technique
through extensive circuit-level simulations and by comparing it to the approach without
node reduction.
The rest of this paper is organized as follows. In Section II, briefs about hardware
trojans and their basic classification. In Section III, the basic idea of the proposed approach is
explained including the weighted input pattern generation algorithm and detection algorithm is
discussed in detail in Section IV. Section V shows the simulation results using the proposed
method, and finally Section VI concludes this paper.
2. Motivation
Design methodologies can assist Trojan detection by employing various testing
techniques. Since an adversary always prefers to use rare internal node conditions to develop a
Trojan, an approach that increases the node activity can facilitate in improving Trojan detection
coverage. An often used strategy is to exercise all functions with critical patterns, instead of the
simulation of thorough set of patterns which can be expensive. However, the definition of crucial
patterns usually depends on the designer’s heuristics [6]. In this paper, the input stimuli is
selected in such a way that it increases the transition probability of all the nodes. By increasing
the transition probability, it is expected that in case of presence of Trojan, its activation time will
decrease and its effect can be seen during the detection phase. Hence, weighted input
probability is generated so as to achieve this condition [7].
For faster and accurate response analysis, observation of internal signals of a circuit
becomes necessary. Controllability, observability and transition probability at each node are few
methods used for observing internal signals [8]. Controllability for a digital circuit is outlined
as the problem of setting a selected logic signal to 0 or 1. Observability for a digital circuit is the
measure of observing the state of a logic signal [8,9]. The controllability and observability
measures are helpful as they quantify the difficulty in setting and observing internal signals of a
circuit [10,11]. Probability defines the risk of obtaining logic 0 or logic 1 on any node of the given
circuit. High probability suggests a high circuit activity at that node. The main advantage of
those techniques is that they involve topological analysis, however no test vectors [12]. These
are static sort of analysis and that they have linear quality, this technique succeeds if the input
test stimuli improves the corresponding node activity. Improving the circuit activity increases the
triggering action and hence increases the chances of observing variations at the output.
Goldenchip free HT detection was discussed in [13], in which power metric based HT detection
and diagnosis is done without referring to any reference chip. This nullifies the effect of process
variation and the reduced power measurements.
3. Proposed Work
Hardware Trojan detection relies on a threat model that explains a specific Trojan
activity within the circuit. Payload is the circuit stricken by the Trojan and trigger initiates the
unwanted change in the circuit activity. The inputs to the trigger is the original nodes with low
circuit activity. Therefore, this is the basis of the algorithm formulated. The rarely triggered
nodes are targeted and the weighted input probability list is generated which increases the
activity at these nodes. Figure 1 gives the brief idea of how Trojan detection is carried out in this
work.
3.1. Generation of Weighted Input Probability List
For activation of Trojans, the transition probability at each node should be increased.
Transition probability threshold ( 𝑇𝑃𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑) is determined as a tradeoff between speed and
efficiency of the detection algorithm. In order to enhance the circuit activity of candidate node n,
whose value is more than 𝑇𝑃𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑, the distinction between its signal chance being 1 and 0
needs to be reduced, so as to realize this, the input probability of the first input that
encompasses a higher influence on the circuit activity of the candidate node will be increased.](https://image.slidesharecdn.com/136812-200821075910/85/Detection-of-Malicious-Circuitry-Using-Transition-Probability-Based-Node-Reduction-Technique-2-320.jpg)
![TELKOMNIKA ISSN: 1693-6930
Detection of Malicious Circuitry Using Transition Probability Based Node ... (Nirmala Devi M)
575
Figure 1. Transition Probability Based Node Reduction Technique
Consider the circuit in Figure 2(a), with 0.5 as the input probability, the transition
probability for all the nodes is calculated using the formulae from Table 1. At node o[2], the
transition probability obtained is 0.234. To increase this value, the transition probability of the
corresponding primary input i.e a[2] is increased to 0.7, which results to 0.249 as the increased
transition probability of node o[2].
Figure 2(a). c17 ISCAS’85 Benchmark circuit
with Initial Transition probability
Figure 2(b). c17 ISCAS’85 Benchmark circuit
with Reduced Transition probability
Table 1. Formulae for Evaluating Transition Probability of Basic Logic Gates
Logic Gate TransitionProbability(TP = P0 x P1)
AND (1-PAPB)x (PAPB)
OR (1-PA)(1-PB)x (1-(1-PA)(1-PB))
NOT (1-PA) x PA
NAND (PAPB)(1-PAPB)
NOR (1-(1-PA)(1-PB))x (1-PA)(1-PB)
XOR (1-(PA+PB-2PAPB))x (PA+PB-2PAPB)
Consider the ISCAS’85 benchmark circuit c1355, it has 41 primary inputs, 32 primary
outputs and total number of gates is 546. When the input transition probability is 0.5, the
minimum circuit activity is noted as 0.0025. After iteratively increasing the probability list, such
that the minimum circuit activity is greater than or equal to the threshold value, it is observed
that the minimum circuit activity can be increased to 0.006 without the need of extra hardware.
This shows that the circuit activity can be increased upto 63.42%.
Step 1 : Levelization of the test circuit.
Step 2 : Transition probability calculation at each node.
Step 3 : Determination of minimum transition probability when input probability is 0.5.
Step 4 : Determination of maximum value to which transition probability can be
increased hence resulting in weighted input probability list.
Step 5 : Calculation of optimum transition probability which can be used as threshold
value.
Step 6 : Acquiring rarely triggered nodes.
Step 7 : Segmentation of the circuit taking radius equal to 1.
Step 8 : Equation formulation and solving based on the leakage power analysis of the
segments obtained.](https://image.slidesharecdn.com/136812-200821075910/85/Detection-of-Malicious-Circuitry-Using-Transition-Probability-Based-Node-Reduction-Technique-3-320.jpg)
![TELKOMNIKA ISSN: 1693-6930
Detection of Malicious Circuitry Using Transition Probability Based Node ... (Nirmala Devi M)
577
identified as rarely triggering nodes in the circuit [14]. Further for Trojan detection, power
analysis is considered wherein the presence of Trojan will show a considerable difference in the
leakage power.
During segmentation, the rarely triggered node is considered as the centre and then by
taking radius as 1, i.e. one logic gate, segments are created and the common gate which has
the rarely triggered node as input is considered as the overlapping gate. Figure 3 is used to
illustrate the above segmentation procedure wherein the Trojan present at the rarely triggered
node will be included in the segment as shown. A change in coefficients is seen in the equations
which are formulated based on the input stimuli given to the segment under observation. The
coefficients are predetermined as it only depends on the gate and the input given to the gate.
The output value of each equation is the overall leakage power of the segment for the
corresponding input stimuli. The characterization of a single gate overlapping in multiple
segments helps to observe change in variable ouput due to presence of malicious circuitry. In
each round of the diagnosis, multiple segments based on number of inputs to the gate with
rarely triggered node as the overlapping gate is characterized. Then, the values are tallied for
various segments. The one that features a considerable distinction compared to the other
values is assumed to be infected [15].
Figure 3. Segmentation Around the Rarely Triggered Node
4. Simulation Results
For various benchmark circuits with different number of controllable nodes, by changing
the transitional probability threshold, the rarely triggered nodes are obtained for each case. It is
observed that if the threshold is too low then many critical nodes are not considered therefore
reducing the system efficiency but the system speed increases whereas if the threshold is very
high then nodes under observation increases which increases system efficiency but system
speed decreases. Hence, selection of transition probability is a tradeoff between efficiency and
speed of the system.
Figure 4. Number of Critical Nodes that Need to be Considered for Varying 𝑇𝑃𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑](https://image.slidesharecdn.com/136812-200821075910/85/Detection-of-Malicious-Circuitry-Using-Transition-Probability-Based-Node-Reduction-Technique-5-320.jpg)
![TELKOMNIKA ISSN: 1693-6930
Detection of Malicious Circuitry Using Transition Probability Based Node ... (Nirmala Devi M)
579
References
[1] M Tehranipoor. Trustworthy Hardware: Trojan Detection and Design-for-Trust Challenges. IEEE
Computer Society. 2011; 44(7): 66-74.
[2] RS Chakraborty. Hardware Trojan: Threats and Emerging Solutions.in IEEE International High Level
Design Validation and Test Workshop, San Francisco, USA. 2009.
[3] M Abramovici. Integrated Circuit Security - New Threats and Solutions. in CSIIR Workshop. 2009.
[4] RS Ranjani, MN Devi. Malicious Hardware Detection and Design for Trust: an Analysis.
Elektrotehniski Vestnik. 2017: 7-16.
[5] M Tehranipoor. A Survey of Hardware Trojan Taxonomy and Detection. IEEE Design & Test of
Computers. 2010; 27(1): 10-25.
[6] B Zhou. Cost-efficient Acceleration of Hardware Trojan Detection through Fan-Out Cone Analysis
and Weighted Random Pattern Technique. IEEE Transactions on Computer-Aided Design of
Integrated Circuits and Systems. 2016; 35(5): 792-805.
[7] R Chakraborty, S Ranjini, N Devi. A Flexible Online Checking Technique to Enhance Hardware
Trojan Horse Detectability by Reliability Analysis . IEEE Transactions on Emerging Topics in
Computing. 2017; 5: 260-270.
[8] M Bushnell. Essentials of Electronic Testing for Digital, Memory and Mixed-Signal VLSI. New York.
Kluwer Academic Publishers. 2006.
[9] H Salmani. COTD: Reference-Free Hardware Trojan Detection and Recovery Based on
Controllability and Observability in Gate-Level Netlist. IEEE Transactions on Information Forensics
and Security. 2016; 12(2): 338- 350.
[10] L Kim. A Trojan-resistant System-on-chip Bus Architecture. in Intl. Conf. on Military Communication.
2009.
[11] X Wang. Detecting Malicious Inclusions in Secure Hardware: Challenges and Solutions. in IEEE Int.
Workshop Hardware-Oriented Security Trust (HOST). 2008.
[12] M Potkonjak. Hardware Trojan horse detection using gate-level characterization. in IEEE Design
Automation Conference. 2009.
[13] RS Ranjani, MN Devi. Golden-chip Free Power Metric based Hardware Trojan Detection and
Diagnosis. Far East Journal of Electronics and Communications. 2017; 17: 517-530.
[14] H Salmani. New design strategy for improving hardware Trojan detection and reducing Trojan
activation time.in IEEE International Workshop on Hardware-Oriented Securityand Trust, Francisco,
CA, USA. 2009.
[15] S Wei. Self-Consistency and Consistency-Based Detection and Diagnosis of Malicious Circuitry.
IEEE Transactions on Very Large Scale Integration (VLSI) Systems. 2014; 22(9): 1845-1853.](https://image.slidesharecdn.com/136812-200821075910/85/Detection-of-Malicious-Circuitry-Using-Transition-Probability-Based-Node-Reduction-Technique-7-320.jpg)
Detection of Malicious Circuitry Using Transition Probability Based Node Reduction Technique
- 1. TELKOMNIKA, Vol.16, No.2, April 2018, pp. 573~579 ISSN: 1693-6930, accredited A by DIKTI, Decree No: 58/DIKTI/Kep/2013 DOI: 10.12928/TELKOMNIKA.v16i2.6812 573 Received November 29, 2017; Revised February 18, 2018; Accepted February 28, 2018 Detection of Malicious Circuitry Using Transition Probability Based Node Reduction Technique Nirmala Devi M*, Irene Susan Jacob, Sree Ranjani R, Jayakumar M Department of Electronics and Communication Engineering, Amrita School of Engineering, Coimbatore, Amrita Vishwa Vidyapeetham, Amrita University, India *Corresponding author, e-mail: m_nirmala@cb.amrita.edu Abstract In recent years, serious concerns have been raised against the tampering of integrated circuits due to outsourcing of circuits for fabrication. It has led to the addition of malicious circuitry known as Hardware Trojan. In this paper, a transition probability based node reduction technique for faster and efficient Hardware Trojan (HT) detection has been attempted. In the proposed method, the fact that the least controllable and observable nodes or the nodes with least transition probability are more vulnerable as Trojan sites is taken into consideration. The nodes that have lesser activity than the threshold are the candidate nodes. At each candidate node, segmentation is done for further leakage power analysis to detect the presence of Trojans. Experimental results observed on ISCAS’85 and ISCAS’89 benchmark circuits illustrate that the proposed work can achieve remarkable node reduction upto 78.81% and time reduction upto 58.7%. It was also observed that the circuit activity can be increased by varying the input probability. Hence, for further reduction in the Trojan activation time, the weighted input probability was obtained. Keywords: Hardware security, Hardware trojans, Transition probability, Weighted input probability, Segmentation Copyright © 2018 Universitas Ahmad Dahlan. All rights reserved. 1. Introduction The increase in the complexity of the modern electronics has led to an increase in the involvement of various firms from around the globe in each phase of the electronics supply chain. Outsourcing has increased the integrated circuit vulnerability to various malicious activities and alterations. These modifications termed as Hardware Trojan(HT) has raised genuine concerns with respect to conceivable dangers to military frameworks, budgetary foundations, transportation security, and healthcare appliances. HTs when introduced in a system, may affect the life span of the system, change the functionality or system specification, leak confidential information or led to denial of service [1-4]. Detection of such malicious alterations is extremely difficult, due to many reasons. To start with, given the vast number of soft, firm, and hard IP centers utilized as a part of SoCs, and in addition the high multifaceted nature of today's IP blocks, distinguishing a little malignant change is extremely difficult. Second, Trojan circuits, by configuration, are normally initiated under certain conditions, which make them inactive and hard to be recognized utilizing random or functional stimuli. Third, tests used to identify fabricating shortcomings, for example, stuck -at and delay faults can not ensure detection of Trojans [5]. Considering these facts, detection of Trojans utilizing basic investigation would be ineffective. An efficient method to accelerate the HT detection is the need of the hour and the same has been proposed in this work. The method is based on the generation of weighted input patterns to facilitate the activation of the HTs. Moreover, transition probability analysis node reduction has been attempted in the circuit for faster detection of HTs the fact that the least controllable and observable nodes or the nodes with least transition probability are perfect trigger nodes for adding Trojans. Hence by considering only these nodes, the Trojan detection process can be accelerated with the same efficiency as the case wherein all the nodes are under observation. For weighted input pattern generation, an iterative procedure is followed such that the transition probability at each node should be equal to or greater than the user- defined threshold. The main contributions of this paper can be listed as follows:
- 2. ISSN: 1693-6930 TELKOMNIKA Vol. 16, No. 2, April 2018 : 573 – 579 574 a. Generation of weighted input probability set, which reduces Trojan activation time. b. Efficient and high speed Trojan detection technique by reducing the nodes of observation. c. Validation of the effectiveness of Transition probability based Node reduction technique through extensive circuit-level simulations and by comparing it to the approach without node reduction. The rest of this paper is organized as follows. In Section II, briefs about hardware trojans and their basic classification. In Section III, the basic idea of the proposed approach is explained including the weighted input pattern generation algorithm and detection algorithm is discussed in detail in Section IV. Section V shows the simulation results using the proposed method, and finally Section VI concludes this paper. 2. Motivation Design methodologies can assist Trojan detection by employing various testing techniques. Since an adversary always prefers to use rare internal node conditions to develop a Trojan, an approach that increases the node activity can facilitate in improving Trojan detection coverage. An often used strategy is to exercise all functions with critical patterns, instead of the simulation of thorough set of patterns which can be expensive. However, the definition of crucial patterns usually depends on the designer’s heuristics [6]. In this paper, the input stimuli is selected in such a way that it increases the transition probability of all the nodes. By increasing the transition probability, it is expected that in case of presence of Trojan, its activation time will decrease and its effect can be seen during the detection phase. Hence, weighted input probability is generated so as to achieve this condition [7]. For faster and accurate response analysis, observation of internal signals of a circuit becomes necessary. Controllability, observability and transition probability at each node are few methods used for observing internal signals [8]. Controllability for a digital circuit is outlined as the problem of setting a selected logic signal to 0 or 1. Observability for a digital circuit is the measure of observing the state of a logic signal [8,9]. The controllability and observability measures are helpful as they quantify the difficulty in setting and observing internal signals of a circuit [10,11]. Probability defines the risk of obtaining logic 0 or logic 1 on any node of the given circuit. High probability suggests a high circuit activity at that node. The main advantage of those techniques is that they involve topological analysis, however no test vectors [12]. These are static sort of analysis and that they have linear quality, this technique succeeds if the input test stimuli improves the corresponding node activity. Improving the circuit activity increases the triggering action and hence increases the chances of observing variations at the output. Goldenchip free HT detection was discussed in [13], in which power metric based HT detection and diagnosis is done without referring to any reference chip. This nullifies the effect of process variation and the reduced power measurements. 3. Proposed Work Hardware Trojan detection relies on a threat model that explains a specific Trojan activity within the circuit. Payload is the circuit stricken by the Trojan and trigger initiates the unwanted change in the circuit activity. The inputs to the trigger is the original nodes with low circuit activity. Therefore, this is the basis of the algorithm formulated. The rarely triggered nodes are targeted and the weighted input probability list is generated which increases the activity at these nodes. Figure 1 gives the brief idea of how Trojan detection is carried out in this work. 3.1. Generation of Weighted Input Probability List For activation of Trojans, the transition probability at each node should be increased. Transition probability threshold ( 𝑇𝑃𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑) is determined as a tradeoff between speed and efficiency of the detection algorithm. In order to enhance the circuit activity of candidate node n, whose value is more than 𝑇𝑃𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑, the distinction between its signal chance being 1 and 0 needs to be reduced, so as to realize this, the input probability of the first input that encompasses a higher influence on the circuit activity of the candidate node will be increased.
- 3. TELKOMNIKA ISSN: 1693-6930 Detection of Malicious Circuitry Using Transition Probability Based Node ... (Nirmala Devi M) 575 Figure 1. Transition Probability Based Node Reduction Technique Consider the circuit in Figure 2(a), with 0.5 as the input probability, the transition probability for all the nodes is calculated using the formulae from Table 1. At node o[2], the transition probability obtained is 0.234. To increase this value, the transition probability of the corresponding primary input i.e a[2] is increased to 0.7, which results to 0.249 as the increased transition probability of node o[2]. Figure 2(a). c17 ISCAS’85 Benchmark circuit with Initial Transition probability Figure 2(b). c17 ISCAS’85 Benchmark circuit with Reduced Transition probability Table 1. Formulae for Evaluating Transition Probability of Basic Logic Gates Logic Gate TransitionProbability(TP = P0 x P1) AND (1-PAPB)x (PAPB) OR (1-PA)(1-PB)x (1-(1-PA)(1-PB)) NOT (1-PA) x PA NAND (PAPB)(1-PAPB) NOR (1-(1-PA)(1-PB))x (1-PA)(1-PB) XOR (1-(PA+PB-2PAPB))x (PA+PB-2PAPB) Consider the ISCAS’85 benchmark circuit c1355, it has 41 primary inputs, 32 primary outputs and total number of gates is 546. When the input transition probability is 0.5, the minimum circuit activity is noted as 0.0025. After iteratively increasing the probability list, such that the minimum circuit activity is greater than or equal to the threshold value, it is observed that the minimum circuit activity can be increased to 0.006 without the need of extra hardware. This shows that the circuit activity can be increased upto 63.42%. Step 1 : Levelization of the test circuit. Step 2 : Transition probability calculation at each node. Step 3 : Determination of minimum transition probability when input probability is 0.5. Step 4 : Determination of maximum value to which transition probability can be increased hence resulting in weighted input probability list. Step 5 : Calculation of optimum transition probability which can be used as threshold value. Step 6 : Acquiring rarely triggered nodes. Step 7 : Segmentation of the circuit taking radius equal to 1. Step 8 : Equation formulation and solving based on the leakage power analysis of the segments obtained.
- 4. ISSN: 1693-6930 TELKOMNIKA Vol. 16, No. 2, April 2018 : 573 – 579 576 Step 1: Computing weighted input probability list such that all the nodes have transition probability greater than 𝑇𝑃𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑 3.2. Node Reduction Algorithm for Trojan Detection Step 2: Using the rarely triggered nodes, the presence of Trojan will be detected In this work, the transition probability at each node is considered for identifying the least controllable and observable nodes i.e. rarely triggered nodes. The rarely triggered nodes come further in use while segmentation is done. Segmentation is implemented around the nodes
- 5. TELKOMNIKA ISSN: 1693-6930 Detection of Malicious Circuitry Using Transition Probability Based Node ... (Nirmala Devi M) 577 identified as rarely triggering nodes in the circuit [14]. Further for Trojan detection, power analysis is considered wherein the presence of Trojan will show a considerable difference in the leakage power. During segmentation, the rarely triggered node is considered as the centre and then by taking radius as 1, i.e. one logic gate, segments are created and the common gate which has the rarely triggered node as input is considered as the overlapping gate. Figure 3 is used to illustrate the above segmentation procedure wherein the Trojan present at the rarely triggered node will be included in the segment as shown. A change in coefficients is seen in the equations which are formulated based on the input stimuli given to the segment under observation. The coefficients are predetermined as it only depends on the gate and the input given to the gate. The output value of each equation is the overall leakage power of the segment for the corresponding input stimuli. The characterization of a single gate overlapping in multiple segments helps to observe change in variable ouput due to presence of malicious circuitry. In each round of the diagnosis, multiple segments based on number of inputs to the gate with rarely triggered node as the overlapping gate is characterized. Then, the values are tallied for various segments. The one that features a considerable distinction compared to the other values is assumed to be infected [15]. Figure 3. Segmentation Around the Rarely Triggered Node 4. Simulation Results For various benchmark circuits with different number of controllable nodes, by changing the transitional probability threshold, the rarely triggered nodes are obtained for each case. It is observed that if the threshold is too low then many critical nodes are not considered therefore reducing the system efficiency but the system speed increases whereas if the threshold is very high then nodes under observation increases which increases system efficiency but system speed decreases. Hence, selection of transition probability is a tradeoff between efficiency and speed of the system. Figure 4. Number of Critical Nodes that Need to be Considered for Varying 𝑇𝑃𝑡ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑
- 6. ISSN: 1693-6930 TELKOMNIKA Vol. 16, No. 2, April 2018 : 573 – 579 578 The proposed method aims to detect malicious circuitry at the gate-level wherin the malicious circuitry is added to the netlist by the adversary. In order to validate the proposed methodology, the ISCAS’85 and ISCAS’89 benchmarks are chosen, in which HT attacks based on addition of malicious circuitry is targetted. Table 1. Transition Probability Based Node Reduction Technique Implemented on ISCAS’85 Benchmark circuit Circuit # Inputs Without Reduction With Reduction % Node Reduction % Time Reduction# Nodes Time (ms) # Nodes Time (ms) c432 36 153 24.742 42 16.173 72.549 34.633 c499 41 170 21.288 48 13.996 71.764 34.254 c1355 41 514 53.569 112 19.165 78.210 64.224 c1980 33 855 37.517 119 19.491 86.082 48.047 c2670 233 1129 68.059 165 26.597 85.385 60.921 c3540 50 1647 57.449 465 29.021 71.767 49.484 c5315 178 2184 122.865 246 35.488 88.736 71.116 c6288 32 2384 173.861 31 27.485 98.699 84.191 c7552 207 3405 230.432 348 42.982 89.779 81.346 82.553 58.691 Table 2. Transition Probability Based Node Reduction Technique implemented on ISCAS’89 Benchmark Circuit Circuit # Inputs Without Reduction With Reduction % Node Reduction % Time Reduction# Nodes Time (ms) # Nodes Time (ms) s298 32 99 38.695 9 16.595 99.909 57.113 s386 20 146 43.766 59 24.269 59.589 44.548 s444 46 154 47.451 25 25.089 83.766 47.126 s526 46 166 50.407 30 29.533 81.928 41.411 s641 73 337 75.173 53 31.205 84.273 58.489 s713 74 351 75.776 41 30.214 88.319 60.127 s838 29 265 66.799 79 34.862 70.189 47.810 s953 52 366 81.411 249 52.691 31.967 35.277 s1488 20 628 111.211 170 47.176 72.929 57.579 s5378 379 2566 202.383 660 69.493 74.279 65.663 s9234 459 5347 446.241 1208 80.376 77.408 81.988 s15850 1132 9102 787.497 1731 107.568 80.982 86.341 s38584 2737 17677 1026.631 3664 364.775 79.272 64.468 75.062 57.534 5. Conclusion In this work, a faster and efficient solution of HT detection using node reduction technique is developed. It employs segmentation around the nodes with transition probability less than the user-defined threshold. It was concluded that while selecting the user-defined threshold, the transition probability should be set such that it is neither too high nor too low because high transition probability will increase the number of segments which is again time consuming whereas low transition probability will decrease the efficiency of the system due to the non-consideration of some of the critical nodes. The weighted input pattern obtained was used for faster Trojan activation as it increased the transition probability at each node to a value which is higher than or equal to the user-defined threshold. The proposed method was implemented on ISCAS’85 and ISCAS’89 benchmark circuits and it was observed that it can achieve remarkable node reduction upto 82.55% and 75.06% respectively, and time reduction upto 58.69% and 57.53% respectively. Hence the proposed technique gives faster and efficient way of Trojan detection. Acknowledgement This work is funded by the Defence Research and Development Organization (DRDO), New Delhi, “ERIP/ER/1503187/M/01/1582".
- 7. TELKOMNIKA ISSN: 1693-6930 Detection of Malicious Circuitry Using Transition Probability Based Node ... (Nirmala Devi M) 579 References [1] M Tehranipoor. Trustworthy Hardware: Trojan Detection and Design-for-Trust Challenges. IEEE Computer Society. 2011; 44(7): 66-74. [2] RS Chakraborty. Hardware Trojan: Threats and Emerging Solutions.in IEEE International High Level Design Validation and Test Workshop, San Francisco, USA. 2009. [3] M Abramovici. Integrated Circuit Security - New Threats and Solutions. in CSIIR Workshop. 2009. [4] RS Ranjani, MN Devi. Malicious Hardware Detection and Design for Trust: an Analysis. Elektrotehniski Vestnik. 2017: 7-16. [5] M Tehranipoor. A Survey of Hardware Trojan Taxonomy and Detection. IEEE Design & Test of Computers. 2010; 27(1): 10-25. [6] B Zhou. Cost-efficient Acceleration of Hardware Trojan Detection through Fan-Out Cone Analysis and Weighted Random Pattern Technique. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems. 2016; 35(5): 792-805. [7] R Chakraborty, S Ranjini, N Devi. A Flexible Online Checking Technique to Enhance Hardware Trojan Horse Detectability by Reliability Analysis . IEEE Transactions on Emerging Topics in Computing. 2017; 5: 260-270. [8] M Bushnell. Essentials of Electronic Testing for Digital, Memory and Mixed-Signal VLSI. New York. Kluwer Academic Publishers. 2006. [9] H Salmani. COTD: Reference-Free Hardware Trojan Detection and Recovery Based on Controllability and Observability in Gate-Level Netlist. IEEE Transactions on Information Forensics and Security. 2016; 12(2): 338- 350. [10] L Kim. A Trojan-resistant System-on-chip Bus Architecture. in Intl. Conf. on Military Communication. 2009. [11] X Wang. Detecting Malicious Inclusions in Secure Hardware: Challenges and Solutions. in IEEE Int. Workshop Hardware-Oriented Security Trust (HOST). 2008. [12] M Potkonjak. Hardware Trojan horse detection using gate-level characterization. in IEEE Design Automation Conference. 2009. [13] RS Ranjani, MN Devi. Golden-chip Free Power Metric based Hardware Trojan Detection and Diagnosis. Far East Journal of Electronics and Communications. 2017; 17: 517-530. [14] H Salmani. New design strategy for improving hardware Trojan detection and reducing Trojan activation time.in IEEE International Workshop on Hardware-Oriented Securityand Trust, Francisco, CA, USA. 2009. [15] S Wei. Self-Consistency and Consistency-Based Detection and Diagnosis of Malicious Circuitry. IEEE Transactions on Very Large Scale Integration (VLSI) Systems. 2014; 22(9): 1845-1853.