IEEE Consumer Electronics Magazine - July 2017 - 131

chip design) driven through particleswarm-optimization (PSO) exploration.
Figure 13 shows the DFS flow adopted
from [7]. The DFS flow accepts the
library details of each component
imported from third-party vendors,
inputs the application in the form of a
control data-flow graph (CDFG), and
produces an optimized design that
guards against this specific type of Trojan. The DFS flow is described next.
The major block in the DFS flow is
a Trojan detection block that provides
security by first generating a dualmodular redundant (DMR) structure of
the application (CDFG) and then
imposing certain hardware-allocation
rules on the operations. Creating a
DMR structure means generating an
original copy and a duplicate copy of
the application simultaneously. Once
the DMR design is created, then it is
concurrently scheduled using a LIST
scheduling algorithm based on the
data (resource configuration) received
from a PSO-driven design space exploration (DSE).
Subsequently, the Trojan specific
hardware-allocation rules are imposed,
which implies that sister operations of
the original and duplicate copies must be
assigned to the hardware from distinct
vendors. This allocation process can be
accomplished by using multiple techniques; thus, optimization here is also
performed using a PSO-DSE. Once
hardware allocation is performed, then
binding and cost evaluation of this DMR
design is evaluated and fed to the PSODSE block to generate the next resource
configuration using the concept of velocity and position upgradation. A particle
in the PSO-DSE block is initialized as
X i = (R 1, f R D, A v, UF), where R 1 to
RD are the various hardware resource
types, A v indicates the mode of possibilities of achieving distinct vendor-hardware allocation, and UF indicates the
loop unrolling factor (in this case, the
application is a loop-based CDFG). This
process of iteration continues until the
PSO-DSE produces a low-cost optimal
Trojan secured design. Figure 14 shows
the vulnerabilities detectable by the DFS
flow in Figure. 13. More details are
available in [7] and [8].

Cout

4-Bit Binary Adder

A

OR
1

1

B3B2B1B0

0

O

En

A3A2A1A0

1

0v

4-Bit Binary Adder

0
En

Trigger

BCD
Adder

B

O3 O2 O1 O0 Payload

FIGURE 10. The DoS Trojan inserted by an adversary in a BCD adder. When Enable (En) = 0,
the designer assumes that this signal enables the adder block and assumes it to be off.
However, En is the trigger signal of the Trojan, and at En = 0, the Trojan logic gets triggered and produces no value at the output of muxes (B 3 B 2 B 1 B 0) . Thus, (O 3 O 2 O 1 O 0)
yields no output or DoS. The black box BCD adder available to a designer is shown
on the right.

T
(1)
(1)
P
Q
(1)

G5

G1

(1)
S

G2

(1)

G4

(0)

(1) R

G6

G3

(1)

T
(1)

G5

G6

(1)
S

(1)
P

G2
G4

G1

Q
(0)

G3

(1) R

(a)

(b)

FIGURE 11. (a) The NBTI stress on G3 when PQRST = 11111 is applied as the input vector.
(b) NBTI stress on G1 and G4 when PQRST = 10111 is applied as the input vector.

OTHER TROJAN DETECTION
TECHNIQUES
There are two types of Trojan detection
techniques: invasive and noninvasive.
Invasive-type Trojan detection aims to
prevent the insertion of Trojans during
the design or fabrication of a chip. Trojan insertion may be likely in cases
where there is dead space in the chip
layout. Dead space is the only available
space for inserting Trojans, because the
total area of the chip cannot be modified
by a rogue element. If a rogue element
can extract the netlist of the design from
its layout, then with rerouting and better
placement/routing, some dead free space
can be created, which may enable Trojan insertion. Thus, to prevent such a
scenario, the authors in [9] proposed

T
(1)
(1)
P

G5
S
(1)

G2

(1)

G4

G1

Q
(0)

(0)
(1) R

G6

G3

(0)

FIGURE 12. The NBTI stress on G3 when
PQRST = 10111 is applied as the input vector.

obfuscating the design to make it difficult for reverse engineering.
Noninvasive techniques can be categorized into two types: a run-time test
and a test-time test. The run-time test
techniques employ an online monitoring system that tries to detect suspicious
activity during an in-field operation,
JULY 2017

^

IEEE Consumer Electronics Magazine

131
Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - July 2017
