EE244 Project Ideas
Noise Analysis
submitted by Edoardo
Charbon
Shepard [ICCAD96] has proposed a method for noise integrity
evaluation in complex digital circuits by generating a noise graph associated
with the propagation potential of noise through pass transistor logic,
as well as static and dynamic gates. In this approach, a static worst-case
noise signal is propagated through the graph and the error induced in any
of the latches in the circuit is evaluated if the noise instability condition
is reached.
There are a number of drawbacks in this method.
First, experience shows that the most critical noise effects, i.e. the
destructive ones, are local (ranges of one or two gates) and do not originate
from remote noise signals. Second, a number of sources such as the input,
output, Vdd, Vss, substrate, etc. act together in a complex pattern to
produce output noise. For example, noise injected at Vdd will be highly
attenuated when the input is low or high and propagated when the input
is in a metastable state. Third, noise is propagated only through
pass gates, which are generally followed by latches, hence not further
propagated. We propose to develop a graph propagation technique which
applies to settling time as opposed to noise signals. Settling time variations
can be efficiently propagated through gates and latches and the resulting
errors can be appropriately evaluated everywhere in the circuit.
Settling time can be compactly formulated as a function of local variables,
such as cross-talk, supply ripple or glitches using, for example, SPICE.
This approach eliminates most of over- or under-estimations and enables
the user to better understand the real effects of noise and cross-talk
during all design phases.
This work consists of the following tasks:
1. construct a settling time model based on random input
vectors
2. evaluate total settling time in macros using a graph
search
3. compute input vectors for each macro based on wire
delay and settling times
The final algorithm will iterate 1. 2. and 3. until consistency
is reached.
Active IP Watermarking
submitted by Edoardo
Charbon
Employing well-known techniques, it is possible today
to reverse-engineer virtually any IC design, given a sufficient pool of
samples and enough time. Developing techniques for direct Intellectual
Property (IP) protection, if ever possible, is impractical and very expensive.
Recently, a new IP protection paradigm has been proposed
based on a technique known as watermarking. Watermarking, applied to hard
IPs, consists in embedding a coded signature within the intrinsic
structure of the IP. Watermarking, to be effective, needs be (a)
unique and highly resilient; (b) hard to detect, counterfeit or remove;
(c) easy to verify. In addition, the procedure needs be transparent
to the user and cheap, and have little or no impact on performance.
In this research we propose to use such unique features as the constraints
on performance and signal interface and the topology of signal, power and
clock interconnections to encrypt the signature. Resilience is ensured
by the highly redundant character of the signature and the different types
of encryption used for each feature encoding. Low hardware overhead
is guaranteed, provided that the layout solution has a large number of
nearly-equivalent configuration, which is always the case in complex IPs
such as the ones we want to protect. We will prove how detecting and forging
a signature can be made arbitrarily hard, while the procedure can be easily
introduced in a large number of existing optimization algorithms, normally
used in physical design.
Passive IP Watermarking
submitted by Edoardo
Charbon
An alternative watermarking technique, involving no encoding
into the IP, consists of using substrate injection signatures as a fingerprint
for a particular logic circuit given a specific set of input vector sequence.
A spectral estimation technique based on the Burg method is then used to
discriminate the signature in a noisy environment, i.e. where other noise
injecting circuits are operating simultaneously. The proposed method
is useful since it allows to detect the presence of a certain IP in a circuit
without re-design, or any modifications to the chip being tested. The signature
characterization could prove useful also for detection of malfunctions
and to pinpoint the exact source of the problem.
NexSis and Layout Generation
submitted by Edoardo
Charbon
In this project we propose the definition of a syntax
for compact representation of modules and interconnect areas to be efficiently
generated in Cadence’s layout environment. The assignment also includes
the development of a general module generator for modules and interconnect
with fixed internal floorplanning to be integrated in the NexSis environment.