Conventional and Unconventional Hardware Architectures, Process Technology
With the current industry trend for Big Data systems substantially being based on a single technology base, (i.e. Intel Xeon Processors) any specific requirements for Big Data are been generalized against the requirement of the broader market, (i.e. PCIe based interface for IO extensibility). The ability of a solution to be optimized to the specific requirement, whether through the balance of IO, memory and compute, or the specific integration of high speed interfaces directly onto a SoC based device, both the HW, and SW architectures can have a substantial effect on the efficiency and capability of a platform to address the specific Big Data requirements. Likewise, emerging process technology advancements could be considered to reduce the upfront costs associated with the development of application specific solutions for Big Data. Together, the system architectures, and the process technologies available to build solutions targeted to the requirements of Big Data will offer the Big Data community an increased chance of solving their challenges.
The task will be based on three related activity areas. First, process technology is the foundational technology on which all the other components of the system stack rest. As we get closer to the limits of Moore scaling, solutions to process-related challenges translate into economic feasibility problems, these issues will be analyzed in relation with Big Data operating costs. Second, at the conventional architecture level, transistor scaling coupled with power-density problems imply the continuation - for the near to medium term - of the multi- and many-core designs. How these designs will need to be overhauled with new process technologies such as 3D stacking or fast non-volatile memories is of special concern. Third, at the unconventional architecture level, recent accomplishments such as the construction of the first graphene transistor, coupled with challenges such as the so-called dark silicon issue (where not all transistors of a processor could be turned on at a given time due to power/thermal reasons) make this segment a very interesting area for Europe to focus on.