Advanced VLSI Design and Test Laboratory

 

The Advanced VLSI Design and Test Group is involved in a number of projects covering a large spectrum of Microarchitecture, Security, VLSI Design and Test Research.

Computer Architecture: On a broad level, our research interests are in path finding for efficient, resilient, and reliable architectures for the multicore era. Today’s designs are being shaped by the challenges of highly scaled technologies: stability and long-term reliability of devices, power dissipation limits that require dynamic adaptation of processor to computing needs. Our research in this area focuses on how to leverage the hardware and software abstraction layers of today’s systems to tradeoff performance, power and reliability in multicores. Several of the low level details of hardware such as power, thermal, and faults are tightly correlated to the hardware behavior, interactions within the system, and application behavior. The conventional approach to tackling such problems in hardware comes with additional costs and high design complexity. On the other hand, software based techniques are severely limited by observability and controllability into the hardware due to the strict abstraction layers of today’s systems. We propose multiple alternatives for dynamic adaptation of hardware. One such approach is based on a middleware call microvisor. The hardware provides the knobs and controls to monitor the low level details at a fine granular level, whereas, the microvisor manages actions for the threads running on the cores. Insulating management from the traditional software (including the operating system) enables a scalable, flexible and secure system level solution and allows the microvisor to evolve freely. Other alternatives include, dedicated controller for dynamic thread monitoring and rescheduling for both symmetric and asymmetric chip multiprocessors. We are also working on the design and evaluation of a novel hybrid NoC architecture which utilizes separate network topologies for real time and best effort traffic so as to achieve higher QoS while simplifying arbitration.

Security: We have published a number of papers recently on Physically Unclonable Functions (PUF). Our current research focuses on PUF and True Random Number Generators (TRNG). PUF and TRNG circuits rely on underlying device characteristics to generate random response. However, they rely on different properties of underlying circuits. PUF circuits are used for chip authentication. Their responses must be reliable and repeatable, implying that PUF response should be invariant to environmental factors such as voltage and temperature disturbances, yet they must vary from one fabricated device to the other implying that they must rely on physical variations. However, the response of the PUF circuit must not be a simple linear transformation of the physical variations which makes them vulnerable to modeling attacks. If they rely on non-linear transformation of physical variations, such transformations must be sufficiently complex to withstand modeling attacks. Even the verifier needs to be able to withstand cloning attacks. It must not send the same sequence of challenges for authentication, as the correct responses may be memorized by an attacker and used subsequently. Our PUF circuits rely on exploiting lithographic distortions while we design these circuits to be insensitive to voltage and temperature to improve uniqueness while also improving their reliability. Our research in security is expanding to include various forms of attacks and countermeasures. We are also exploring how to use/secure NVRAM for crypto applications. Stay tuned for publications in this area soon.

VLSI CAD, Design Methodologies & Test: Our research in this area includes logic synthesis for emerging devices such as FinFETs, clocking and clock network design, 3D architectures, dynamic pattern generation for worst case power analysis, thermal simulation, lithography simulation, dual pattern lithography, optical proximity correction, lithographic error modeling, yield analysis, circuit reliability analysis, soft-error analysis, modeling of non-rectangular transistors and 3D devices. We have also explored suitability of GPGPUs for CAD algorithms. We have recently published a number of low-cost solutions for improving online error detection, diagnosis and repair that achieve nearly 100% coverage at about 15% energy overhead. We have a rich publication record in this area.