Reconfigurable On-Chip Instrument Access Networks : Analysis, Design, Operation, and Application

Abstract: The constant need for higher performance and more advanced functionality has made the design and manufacturing of modern electronic chips highly demanding. Moreover, the use of smaller transistors in modern chips has increased their sensitivity to aging and faults, hence the need to constantly monitor the correct operation of these chips. To address the challenges and requirements, it has become common to embed extra hardware modules in the chips to assist in the design and manufacturing processes, as well as in monitoring the correct operation of the chips. Such modules, commonly referred to as on-chip instruments, are used through the entire life cycle of the chip, from the early prototyping phase to when the system incorporating that chip becomes operational at the customer's site. The increasing trend in the number and complexity of the on-chip instruments called for methodologies that allow for scalable, fast, and easy access to these instruments. As an alternative to in-house methods, which although effective might be expensive to maintain, two IEEE standards, namely, IEEE Std 1687 and IEEE Std 1149.1-2013, have recently come into existence. These standards provide a common base for describing reconfigurable on-chip instrument access networks, as well as for describing the operation of each embedded instrument by using high-level description languages. Such common base motivates the development of relevant design automation tools, and facilitates the integration of instruments developed by multiple vendors. These standards, however, have left the arising optimization problems in the design and operation of such networks to be addressed by the electronic design automation community. In this thesis, we address some of these optimization problems whose objective is to minimize the instrument access time, i.e., the time it takes to transport data to/from the instruments over reconfigurable on-chip instrument access networks. In particular, we present access time analysis that helps to determine the contributing factors to the access time overhead. Using the analysis, we present methods for design of reconfigurable networks that are optimized with respect to instrument access time. Moreover, to operate such on-chip networks, there is a need for automation tools that translate (retarget) high-level descriptions of instrument access procedures specified at instruments' boundaries, into low-level description languages or bit vectors applicable from the chip's boundary. The reconfigurability of these networks, makes it challenging to perform the retargeting such that the generated vectors are optimized with respect to the time it takes to apply them to the chip. In this thesis, we explore opportunities for optimization in retargeting. In particular, we present a method to assist in optimal bit vector generation, by reducing the solution space without removing the optimal vector from it. Finally, considering the application of on-chip networks in in-field monitoring of the correct operation of chips, we propose a self-reconfiguring network that upon detection of errors, automatically reconfigures itself to reduce the time it takes to identify the faulty resources.