Abstract
With the introduction of FDSOI processes at competitive technology nodes, body biasing on an unprecedented scale was made possible. Body biasing influences one of the central transistor characteristics, the threshold voltage. By being able to heighten or lower threshold voltage by more than 100mV, the very physics of transistor switching can be manipulated at run time. Furthermore, as body biasing does not lead to different signal levels, it can be applied much more fine-grained than, e.g., DVFS. With the state of the art mainly focused on combinations of body biasing with DVFS, it has thus ignored granularities unfeasible for DVFS. This thesis fills this gap by proposing body bias domain partitioning techniques and for body bias domain partitionings thereby generated, algorithms that search for body bias assignments. Several different granularities ranging from entire cores to small groups of standard cells were examined using two principal approaches: Designer aided pre-partitioning based determination of body bias domains and a first-time, fully automatized, netlist based approach called domain candidate exploration. Both approaches operate along the lines of activation and timing of standard cell groups. These approaches were evaluated using the example of a Dynamically Reconfigurable Processor (DRP), a highly efficient category of reconfigurable architectures which consists of an array of processing elements and thus offers many opportunities for generalization towards many-core architectures. Finally, the proposed methods were validated by manufacturing a test-chip. Extensive simulation runs as well as the test-chip evaluation showed the validity of the proposed methods and indicated substantial improvements in energy efficiency compared to the state of the art. These improvements were accomplished by the fine-grained partitioning of the DRP design. This method allowed reducing dynamic power through supply voltage levels yielding higher clock frequencies using forward body biasing, while simultaneously reducing static power consumption in unused parts.
Author and committee
dc:creator, dc:contributor.*- Author
-
- Kühn, Johannes Maximilian
Identifiers
dc:identifier.*- Identifier
- hdl:10900/78961