Non-Volatile FPGA: Towards Efficient Versatile Internet of Things in a Smarter City

Description

In order to keep its leading roles as the world’s financial center, it is essential for HongKong to utilize modern technologies to build a smarter city. FPGA, with excellentscalability and flexibility to customize in both hardware and software, is a promisingtechnology for various smart city and Internet of Things (IoT) applications. However, asthe technology node scales under 20 nm, traditional SRAM-based volatile FPGAs arefacing severe challenges such as high leakage power and process variations. To addressthis problem, non-volatile memory can be alternatively adopted in FPGAs as Look UpTables (LUT) and other memory components to achieve high scalability, small formfactor, and low power consumption.When NVMs are employed in FPGA, existing techniques need to be renovated due to thefollowing reasons. First, traditional FPGA synthesize and reconfiguration techniqueswere optimized for SRAM-based memory components. Therefore, they did not take intoconsideration the inherently NVM characteristics such as read/write asymmetry. Thusthe FPGA synthesize and reconfiguration procedures need to be redesigned to efficientlyutilize NVM. Second, none of existing NVM optimizations are FPGA-specific. Thus,these techniques should be further improved by considering FPGA working patterns inNVM-based FPGAs. Third, different from SRAM, NVMs have the capability of storingmore than one bit in a cell, which is known as Multiple Level Cell (MLC). While itcertainly increases storage density and reduces FPGA size, it also complicates the designand configuration of NVM based FPGA. Thus, techniques to efficiently adopt MLC NVMin FPGA need to be developed.This project aims to fill these gaps and develop a long-lasting, fast-adaptive, reliable,and energy-efficient platform for future self-adaptive systems. Design issues will bestudied with various optimization goals including write operation efficiency andendurance of NV-FPGA memory, reconfiguration cost reduction, access modeling ofNV-FPGA, read optimizations through retention time aware mode selection, as well assystem-level hardware/software co-design. With the proposed tasks, an efficient NVFPGAplatform as well as NVM characteristics aware FPGA synthesis/reconfigurationtechniques will be developed.The success of this project will enable a new class of efficient non-volatile FPGA basedembedded systems that can be utilized to build a smarter city for Hong Kong, with whichwe can provide better environmental monitoring, traffic management, smart buildings,and enhanced public safety. All of these will in turn help Hong Kong to maintain andadvance its role as one of the world’s leading financial center.