Welcome to Zhimin Chen's homepage
| Virtual Secure Circuit on ASIP with Customized Instructions | |
| SES Group, Virginia Tech, Blacksburg, VA | Mar. 2010 - Present |
ASIP with customized instructions is another approach to implement Virtual Secure Circuit (see below for definition). Unlike the dual-core PowerPC system, the ASIP-based approach aims to emulate the dual-rail pre-charge circuit with only one single processor. To fulfill this purpose, half of a processor needs to behave as the direct-rail while the other half as the complementary-rail. Such processor is called 'balanced processor'. We designed a balanced processor by customizing a Leon3 processor. In detail, we added two simple customized instructions, including balanced AND (b_and) and balanced OR (b_or). Together with the balanced programming method based on the bitslicing programming, we have been able to implement the full AES in the protected VSC fashion. With Measurement to Disclosure (MTD) as the quantification approach, the custom-instruction VSC offers 20 times more security. |
|
| Virtual Secure Circuit on the Dual-Core PowerPC System | |
| SES Group, Virginia Tech, Blacksburg, VA | Aug. 2009 - Present |
|
We present a concept, called Virtual Secure Circuit (VSC), to enable the processors to emulate the dual-rail pre-charge circuits with software programming. The objective is to protect the software cryptography against Side-Channel Attacks. VSC on the dual-core PowerPC system is an implementation of this concept, where each PowerPC core is programmed to have the same behavior of one rail of the dual-rail circuit. With 3 techniques, including 1) programming with complimentary instruction pairs, 2) inserting pre-charge instructions, and 3) synchronizing two PPC cores cycle by cycle, we successfully ported the dual-rail pre-charge technique to dual-core software. Experimental results showed that the security of the VSC-based software had been improved by 80 times, which is comparable to the dual-rial pre-charge circuit (e.g. WDDL). Besides the security improvement, another good feature is that our design does not need any modification to the processors. Details can be found in a technical report [pdf]. The design files for the SASEBO-G platform are available upon requests. |
|
| Multi-core eMIPS processor with instruction set extensions | |
| Microsoft Research, Redmond, WA | May 2009 - Aug. 2009 |
| A Multi-core eMIPS processor was built in 4 Virtex 5 FPGAs on BEE3 board together with my mentor Neil Pittman. The multi-core system consists of 8 eMIPS cores with 2 cores in each FPGA. On-chip shared memories are employed for communication within one FPGA while message passing network is used for inter-chip communication. Different from the regular dual-core processors, this multi-core eMIPS processor supports instruction set extensions (rISE). Further research showed that multicore and rISE explores different levels of parallelism. Combining them together gave us higher performance and higher efficiency. In detail, rISE gives multicore more balanced task loads; multicore offers rISE a higher-level view which can be used to increase rISE's area efficiency. For more details, please refer to a technical report available at http://research.microsoft.com/apps/pubs/?id=102029. A research paper is also coming up in FPGA2010. | |
| Parallel programming of Montgomery multiplication on multicore systems | |
| SES Group, Virginia Tech, Blacksburg, VA | Jan. 2009 - May. 2009 |
| A scalable parallel programming scheme for Montgomery multiplication was devised for multicore systems. We call it parallel Separated Hybrid Scanning (pSHS). pSHS offers stable performance, high portability, high throughput and low latency over different multicore platforms. This makes pSHS a good candidate for public-key software implementations. A high-light point of pSHS is that it can accommodate several clock cycles communication latency among processing cores. Another high-light point is that pSHS not only increases the throughput but also reduces the latency. A multicore prototype was built to demonstrate pSHS with at most 8 MicroBlaze cores in a Virtex 5 FPGA. A paper on this has been published in DATE2010 [pdf]. | |
| High precision digital thermal sensors | |
| SES Group, Virginia Tech, Blacksburg, VA | Oct. 2008 - Jan. 2009 |
| Do you believe that IC chips have skin that can sense a human being's finger touch? Is it possible that two circuit modules on one chip can cummunicate without wire connections? The answer to the above questions is YES! We designed a digital thermal sensor with high precision. With this technique, an FPGA can correctly and stably detect the finger touches on its package. Two on-cihp circuit modules can communication via a thermal channel enabled by a pair of heat generator and a thermal sensor. The underlying principle is very simple: circuits become slower as the temperature increases. Our digital thermal sensor is based on this phenomenon. A ring oscillator (RO) is build with digial NOT gates. A counter attached to the RO helps to monitor its frequency. Finally, the sensor's precision is increased with a very simple signal processing process. This design won the 4th place in CSAW2008 embedded system contest with honorable mention. For more details, please refer to the paper presented at ISVLSI2009 [pdf] and a technical report for CSAW2008 [pdf]. | |
| Standard hardware interface for HASH functions | |
| SES Group, Virginia Tech, Blacksburg, VA | Oct. 2008 - Jan. 2009 |
| Coming soon ... | |