The Research Of General Parallel Vector Crypto Coprocessor

Cryptograph is one of the most important methods to guarantee the security of information and plays a more and more important role in the applications such as data storage and network communication. As users increase their requirement on data security with the evolvement of computing system, new crypto algorithms are emerging and crypto standards are updating. Security systems or security protocols utilize kinds of crypto algorithms to satisfy different security demands. These requirements force the crypto processing system to support many crypto algorithms and still maintain sufficient performance.The general purpose processor use instructions to perform different crypto programs, while the performance it can achieve is poor. The ASICs improve the performance of some crypto algorithms but are not flexible. The reconfigurable hardware is a trade-off between general purpose processor and ASIC, but it need more time to be mature.Vector processing is a technique to process large amount of data with intensive computing. Using vector processing technique, micro-processor can develop more Data Level Parallelism (DLP) in crypto processing and improve the performance. Based on the vector processing technique and the characteristics of crypto processing, we carried out our study on the design of efficient and flexible general crypto processor architecture. The main work and research fruit is as follow.1. Based on the characteristics of crypto processing, the Parallel Vector Crypto Processing Model (PVCPM) is promoted following the vector processing principle. The PVCPM is composed of Vector Processing Model (VPM), Control and Management Model (CMM), Interconnection Model (ICM) and Memory Access Model (MAM). In this model, multiple VPMs can work cooperatively or independently. The PVCPM use hidden register transfer mechanism to support limited vector linking. It uses centered instruction dispatching, distributed instruction issuing and executing methods to explore Instruction Level Parallelism. With multi-rail buffers in MAM, the PVCPM can satisfy different data access pattern in crypto processing. The PVCPM is the fundamental to research the crypto-processor architecture, instruction set architecture and the mechanism of instruction executing.2. Based on the PVCPM and the characteristics of crypto operations, we proposedthe Parallel Vector Crypto Operation Set (PVCOS). The PVCOS describes the essentials' specialty of crypto algorithms and expresses the main mechanics within PVCPM. Based on the PVCPM and the PVCOS, the Parallel Vector Crypto Processor (PVCP) architecture is introduced and the Instruction Set Architecture of PVCP is depicted.3. The PVCP were designed and the key mechanisms of all components are analyzed in this dissertation. The dispatcher control algorithm used in Control and Management Unit (CMU) can translate outer instructions into inner instructions and dispatch them quickly The hybrid dispatch algorithm can decreases the register transfers between Vector Processing Units (VPUs) and avoids the 'hot' spot by taking into account of registers' assignment and busy/idle state of VPUs. Inter-Connection Unit (ICU) use three-step handshake protocol to correctly transfer data and perform simple instructions. In VPU, the architecture and data hazard can be determined by checking functional units' occupation and registers' usage. The hazard determination algorithm is simple and can ease the hardware design. In MAU, multi-buffers architecture was designed to satisfy different data access pattern in crypto processing.4. The Verilog description and functional simulation of our design based on Xilinx FPGA were done. The simulation results proved its correctness and efficiency. The experiment and evaluation results indicate that when block ciphers were performed, PVCP is 1.22~5.08 times faster than the general high performance microprocessors and is 0.99-2.72 times faster than other crypto processors. It also can carry out block ciphers in fewer cycles than other implementations do. When public-key ciphers were performed, PVCP can achieve comparative performance as the general high performance microprocessors and exceed some hardware implementations.Conclusion can be drawn that the work in this dissertation made a good explore in micro-processor architecture design methodology, and the utilization of vector processing mechanism makes the system flexible and efficient in crypto processing. The results showed that the PVCP can guarantee equal or higher performance than the general microprocessor and some specific crypto processors, which makes it suitable to be used as the core of the security system.