Defense Event

Achieving Correctness in Large Interconnect Systems

Rawan Abdel Khalek

Tuesday, November 10, 2015
09:00am - 11:00am
3725 Beyster Bldg.

Add to Google Calendar

About the Event

In today's semi-conductor industry, large chip-multiprocessors and systems-on-chip are being developed, integrating a large number of components on a single chip. The sheer size of these designs and the intricacy of the communication patterns they exhibit have propelled the development of network-on-chip (NoC) interconnects as the basis for the communication infrastructure in these systems. However, faced with the interconnect's growing size and complexity, several challenges hinder its effective validation. During the interconnect's development, the functional verification process relies heavily on the use of emulation and post-silicon validation platforms. However, detecting and debugging errors on these platforms is a difficult endeavor due to the limited observability, and in turn the low verification capabilities, they provide. Additionally, with the inherent incompleteness of design-time validation efforts, the potential of design bugs escaping into the interconnect of a released product is also a concern. This dissertation provides solutions to enable the development of functionally correct interconnect designs. I first address the challenges encountered during design-time verification efforts, by providing two complementary mechanisms that allow emulation and post-silicon verification frameworks to capture a detailed overview of the functional behavior of the interconnect. My first solution re-purposes the contents of in-flight traffic to log debug data from the interconnect's execution. This approach enables the validation of the interconnect using synthetic traffic workloads, while attaining over 80% observability of the routes followed by packets and capturing valuable debugging information. I also develop an alternative mechanism that boosts observability by taking periodic snapshots of execution, thus extending the verification capabilities to run both synthetic traffic and real-application workloads. Moreover, I develop error detection and recovery solutions to address the threat of design bugs escaping into the interconnect's runtime operation. These runtime techniques can overcome communication errors without needing to store replicate copies of all in-flight packets, thereby achieving correctness at minimal area costs.

Additional Information

Sponsor(s): Valeria Bertacco

Open to: Public