Reconfigurable Computers in Space: Problems, Solutions and Future Directions

1. Reconfigurable Computers in Space: Problems, Solutions and Future Directions Neil W. Bergmann, Anwar S. DawoodCRC for Satellite SystemsQueensland University of TechnologyGPO Box 2434, Brisbane 4001Australia Phone: +61-7-3864-2785, Fax: +61-7-3864-1516 E-mail: n.bergmann@qut.edu.au, a.dawood@qut.edu.auWWW: http://www.crcss.qut.edu.au/

2. Summary Interest in Reconfigurable Computing (RC) has recently spread to those interested in space missions. Although RC has sparked much interest in the general computing community, it has yet to demonstrate “killer app” status for any terrestrial applications. However, we believe that there are several compelling arguments about why RC is an excellent match to the requirements of space missions. This paper: • Describes these arguments, • Looks at characteristics of the space environment, • Looks at possible solutions to the use of RC in space, and • Looks at possible long-term research directions.

3. Why can space be a “killer app” for Reconfigurable Computing? • After launch, unmanned spacecraft electronics are generally unavailable for physical upgrade or repair. RC technology allows new hardware circuits to be uploaded via a radio link. • New circuit configurations can overcome design faults, allow improved processing algorithms to be uploaded, or change system functionality in response to changing mission requirements. • The same circuitry can be used with different configurations at different stages of a mission, reducing weight and power requirements. • If part of an FPGA fails, then circuitry can be reprogrammed to make use of remaining functional portions of the chips. • Use of FPGAs allows generic circuit boards to be designed, which are customised for individual applications. This helps overcome the very high NRE costs associated with small volume spacecraft design. Physical and environmental qualification costs can also be shared across many missions. • In-flight reconfiguration provides additional safety margins for missions with very short lead-times, or for those where mission requirements are not fullt defined at launch.

4. Problematic Aspects of Operating in Space • Ionising radiation causes soft-errors in the static RAM cells used to hold programming information in FPGAs.Longer-term ionising radiation causes hard-errors in the electronic circuitry. • Radio-links to spacecraft are often low bandwidth and high error-rate. This is not a good match to the relatively large configuration files of order 1 Mbit required for modern FPGAs. • Limited on-board memory restricts the number of different configurations that can be stored. Uploaded alternative configurations stored in EEPROM are also susceptible to radiation-induced errors.

5. Short-term solutions for configuration errors • FPGAs generally allow the configuration bit-stream to be read back to check for errors. The simplest FPGA-configuration error-detection technique simply examines the readback bitstream and compares it to the correct bitstream (or alternatively compares the CRC signature of the read-back stream to the desired). Correction is by reloading the FPGA configuration. • Triple-redundancy voting circuits allow faulty FPGA circuits to be switched out and reprogrammed while the system is still operating.

6. Short-term solutions for configuration management • Techniques to assist with uploading new configurations aim to reduce configuration file sizes for storage and transmission. • Techniques include specialised compression techniques, and differential configuration formats (relative to an on-board default configuration). • CRC checks are necessary for on-board monitoring of configuration file integrity. • For deep space missions, error correcting codes are desirable.

7. Long-term solutions requiring special “space-friendly” FPGAs • Space-friendly FPGAs should provide on-chip configuration error-detection and/or correction circuitry which operates continuously and unobtrusively. • Techniques will need to be developed to identify permanently faulty logic blocks within an FPGA, most likely by loading a special set of diagnostic configurations. • Techniques are needed to allow existing circuit designs to be reconfigured on-board the spacecraft to avoid faulty logic cells. This is impractical with current generation place-and-route software. • There is much scope for research into error-detecting or fault-secure logic circuit designs for FPGAs, based on a new “single configuration bit-flip” error model. These internally redundant logic gate designs could build strong fault tolerance into existing FPGA chips. • A single chip or MCM combination of FPGA, microcontroller, flash memory configuration store, and digital and analog I/O circuitry would greatly reduce space mission weight and cost.

8. Adaptive Instrument Module Payload • This is being designed for the experimental FedSat LEO, due for launch in late 2001. • This experiment provides a vehicle for validating our research ideas. • A small (1kg) payload consisting of • Microcontroller with RS422 communications • SRAM-based FPGA (Xilinx X4062) • RAM, PROM, Flash Memory storage • Adaptive Instrument Port

9. Conclusions • Reconfigurable computing has many advantages for space applications, and is an excellent match to new directions in low-cost, flexible space missions. • Existing FPGA architectures have significant operational issues (eg radiation induced errors) and management issues (such as configuration management). • In the short term these issues can be identified and ameliorated by additional external circuit and system techniques • In the long-term, these techniques need to be included as part of the FPGA chip design. Overall, there is potential for space-based computing to be a “killer app” for reconfigurable computing technology.