Behavioral Language Converter to Improve HLS Quality of Results
High Level Synthesis (HLS) is one of the most emerging fields of research that is helping the industries to reduce the time-to-market constraint. With the advancement of the HLS, till now approximately 100 companies have adopted this for IP(Intellectual Property) design. The higher we go in abstraction level, the lesser time it takes to design and verify an IP or a design. It will take tremendous amount of time in comparison to HLS, to implement an IP in verilog or VHDL. This thesis focuses on improving the Quality of Result of HLS by an automation flow with added security(Obfuscation) to the IP. If it’s Hardware Description Language (HDL), the input type does not have any impact in the Quality of Result, because the architecture are being defined in the low level HDL coding. In HLS, as we are coding in higher abstraction, the architecture are being defined by the HLS tools. This might vary based on the optimization techniques that the tools follow as a front end parser, which converts the input to a CDFG (Control Data Flow Graph), which in turn is the input to HLS. So, different input languages will end up generating non-identical architecture for the same design, one of them being the best and one of them being the worst. We need to decide which input language should be considered for a particular design, which will generate the best QoR. One of the issues that also needs to be addressed is SECURITY. The IPs must be made secure or encrypted to prevent unlawful usages. Usually, consumers buy IPs for a trial or evaluation purpose for a few days/weeks to verify the efficiency of the IP to their requirements. If the IPs will be made visible to the consumers in trial phase, they won’t need to end up buying it, which will be a huge loss for the vendor. To prevent this kind of scenario, we need to encrypt the IP. This thesis proposed a method “Obfuscation in HLS”, which is the most easy, inexpensive and efficient way to encrypt an IP. However, the QoR(Quality of Result) of HLS is greatly impacted with Obfuscation. This thesis, proposes a unique conversion tool which converts the input language to SystemC from ANSI-C to save time, resources and will also automatically generate which input language to choose for the best QoR while adopting HLS.