A Theory of Reliable Hardware

The goal of the project is to develop hardware that is more reliable. This is achieved by developing a theory of fault-tolerant hardware, i.e. understanding fundamental principles in dealing with both transient and permanent faults of any kind. The developed ideas are then implemented to measure how successful they are in practice.

A main focus of the project is the development of highly reliable, accurate, and efficient clock generation and distribution methods. Traditional designs cannot cope with permanent faults and have limits in scalability, which we address by devising fault-tolerant distributed clocking methods. If successful, this enables faster, better, and cheaper computers. As computers (and computing devices) are omnipresent, this has the potential of large economical benefits.

Within this reporting period, we have made major strides towards a better understanding of metastability-containing circuits. This enables us to devise high-frequency control loops with digital control, but without synchronizers. Avoiding synchronization delay bears the promise of better circuits for a number of problems. As an "unintended" result, this led to better circuits for adaptive voltage control in face of voltage droops. We have filed several patents and are working towards setting up a project that will produces prototype circuits demonstrating the power of our techniques, with the goal to attract industry partners and, ultimately, bring these technologies to market.

The above approach is based on worst-case modeling of metastability propagation. Despite some impossibility/hardness results, we were positively surprised by how far this can be taken. Apart from theoretical insights, we also learned that, intuitively, metastability of storage elements can usually (or almost always?) be effectively masked into late output transitions. This suggests that for the main objectives of the project, devising a "more universal" framework for metastability modeling is not crucial. Overall, we made significant progress with respect to the first objective.

As evidenced by the listed publications, there has been progress on Objective 2 as well; a number of algorithms and general constructions has been obtained, mostly, but not exclusively, in the context of metastability-containing computations.

Concerning Objective 3, no technological impact has been realized so far. Given the long road from theory to a product, this was to be expected in the first reporting period. I would have hoped for industry and practice-oriented researchers to be more willing to collaborate towards products. As this is not the case (yet), I'm planning to set up the aforementioned project with Milos Krstic in order to obtain tangible results that are easier to market.

Apart from the reduced focus on (a) more general model(s) for metastability, there is currently no change in goals or expected results of the project.

Periodic Reporting for period 3 - ToRH (A Theory of Reliable Hardware)

Diese Seite teilen

Herunterladen