Final Report Summary - APEX (Advanced Processor Core for Space Exploration)
The primary objective of the APEX project is to conduct pre-commercial research to lay the foundations and establish the mechanisms for the development of a fault and radiation tolerant ARM processor to be used in the computing systems of future space exploration missions. The project is aimed at understanding and capturing the requirements of future space applications and addressing them by developing, verifying and validating a novel fault-tolerance technique on an ARM CPU with minimal impact on performance and power-efficiency to meet the requirements of NASA’s future space applications. The motivation for this project comes from the performance crisis currently seen in the space field, where space-qualified microprocessor technology (e.g. RAD750 processor) lags commercial devices (e.g. ARM processors) by several generations and fails to meet the near-to mid-term needs of space exploration missions.
In the first year of the project, Dr Iturbe stayed at the NASA Jet Propulsion Laboratory (JPL) to work on the Compositional InfraRed Imaging Spectrometer (CIRIS) as a case-study. Spectroscopy allows scientists for determining the composition of planetary bodies and is of utmost importance in the search for life beyond Earth, which is one of the major objectives of space exploration for the next decades. The CIRIS spectrometer is a promising JPL instrument proposed for deployment in NASA's future missions to Venus, Mars, asteroids, comets, as well as Jupiter's moons Europa, Callisto, Ganymede and Io. It is a lightweight, rugged and compact device, whose major innovation is the fact that spectroscopy data is acquired in the time domain and then translated into the frequency domain. Although this functioning results in higher computation requirements, it enables dealing with the data corruption provoked by radiation hits in the instrument’s photo-detectors that manifest as large current pulses (i.e. significantly greater than the nominal value) in the time domain. In addition of being representative of the computation requirements expected in future space instrument payload systems, the CIRIS spectrometer demands an extremely effective fault-tolerance level to be used in hostile space environments, such as Jupiter’s high radiation belts.
The major outcome of the APEX project so far is a Xilinx Zynq-based System-on-Chip (SoC) platform, the APEX-SoC, which provides a convenient infrastructure for hardware and software based science data processing as well as other non-instrument-related functions, such as the communications with the spacecraft avionics where the instrument is to be integrated as a subsystem. Most of the CIRIS data processing stages have been implemented on the Zynq’s programmable logic (i.e. Xilinx 7-series FPGA fabric), whereas the stages that involve floating-point operations run as software in the Zynq’s processor (i.e. ARM Cortex-A9). The APEX-SoC platform accomplishes a dual approach to cope with harsh radiation in space: a signal processing stage eliminates most of the radiation hit pulses that corrupt the CIRIS data and a set of fault-tolerance features implemented on the Zynq FPGA fabric prevent, detect and handle radiation-provoked malfunctions in the programmable logic. These techniques include the use of one-hot encoding in finite state machines, ECC protection of memories and periodical checks of the correctness of Zynq’s configuration memory. Two instances of the CIRIS data processing stages are used to increase the reliability of the system, simultaneously processing the data delivered by the same photo-detector and repeating the computation if there is a divergence in the results. The two processing stages can also work independently of each other to improve performance by processing data collected from different photo-detectors in parallel. Fault-diagnostic components are also included in the APEX-SoC to detect deviations from the expected instrument functioning, such as stuck-at situations in data transfers, malfunctions of the electromechanical components, die overheat situations or anomalies in the power supply. It is important to note that at this stage of the project we have not implemented any fault-tolerance mechanism on the processor yet, i.e. we rely on using the Cortex-A9 processor in the Zynq.
The CIRIS controller implemented on the APEX-SoC quadruples the performance requirement of the instrument as it is able to process spectroscopy data delivered by up to 100 photo-detectors (762 Mb/s), while the initial requirement was only 25. Besides, the APEX-SoC-based CIRIS controller only consumes around 5W, which is almost 40% less energy of that consumed by an equivalent board based controller. Finally, when irradiating the instrument photo-detector with a 60Co ɣ-ray source (1 rad/sec: approx. 3,400 hits/sec), the controller is able to compute a spectrum with the same spectral content of that corresponding to the non-irradiated spectroscopy data and with a comparable Signal to Noise Ratio (SNR).
Dr Iturbe spent the second and final year of the fellowship at the host institute, ARM, Cambridge in the UK. He focussed on a thorough study of the ARM CPU family to match the requirements of future space applications captured at NASA. Cortex-R5 CPU is identified as the most suitable ARM CPU, and investigated an architecture-level radiation tolerance technique. Dr Iturbe implemented a triplication model at the core level to build a radiation-tolerant Cortex-R5 CPU sub-system. The three Cortex-R5 CPUs are lockstepped sharing the caches and memories that are protected by SECDED code. He set up a fault injection methodology to emulate the radiation effects on the triple-core Cortex-R5 CPU sub-system. The sub-system is also implemented using commercial libraries to measure the performance, area and power efficiency overheads of the radiation-tolerant Cortex-R5 sub-system and compared to the non-radiation tolerant Cortex-R5.
Being a worldwide company with headquarters in Europe and leader of high-performance and energy-efficient mobile and embedded processor market, ARM has a big opportunity to become the leader of high-performance and energy-efficient radiation-tolerant processors. This will open the door to affordable space technology and the creation of a community of hardware and software developers (i.e. researchers and engineers) around it. In the long- term, the space/aerospace community could be integrated into the mainstream ARM eco-system, allowing for sharing the developer skills and expertise in the commercial mobile and embedded systems with the space community to create synergies around the ARM processor technology. This would have a tremendous impact on the European competitiveness.
In the first year of the project, Dr Iturbe stayed at the NASA Jet Propulsion Laboratory (JPL) to work on the Compositional InfraRed Imaging Spectrometer (CIRIS) as a case-study. Spectroscopy allows scientists for determining the composition of planetary bodies and is of utmost importance in the search for life beyond Earth, which is one of the major objectives of space exploration for the next decades. The CIRIS spectrometer is a promising JPL instrument proposed for deployment in NASA's future missions to Venus, Mars, asteroids, comets, as well as Jupiter's moons Europa, Callisto, Ganymede and Io. It is a lightweight, rugged and compact device, whose major innovation is the fact that spectroscopy data is acquired in the time domain and then translated into the frequency domain. Although this functioning results in higher computation requirements, it enables dealing with the data corruption provoked by radiation hits in the instrument’s photo-detectors that manifest as large current pulses (i.e. significantly greater than the nominal value) in the time domain. In addition of being representative of the computation requirements expected in future space instrument payload systems, the CIRIS spectrometer demands an extremely effective fault-tolerance level to be used in hostile space environments, such as Jupiter’s high radiation belts.
The major outcome of the APEX project so far is a Xilinx Zynq-based System-on-Chip (SoC) platform, the APEX-SoC, which provides a convenient infrastructure for hardware and software based science data processing as well as other non-instrument-related functions, such as the communications with the spacecraft avionics where the instrument is to be integrated as a subsystem. Most of the CIRIS data processing stages have been implemented on the Zynq’s programmable logic (i.e. Xilinx 7-series FPGA fabric), whereas the stages that involve floating-point operations run as software in the Zynq’s processor (i.e. ARM Cortex-A9). The APEX-SoC platform accomplishes a dual approach to cope with harsh radiation in space: a signal processing stage eliminates most of the radiation hit pulses that corrupt the CIRIS data and a set of fault-tolerance features implemented on the Zynq FPGA fabric prevent, detect and handle radiation-provoked malfunctions in the programmable logic. These techniques include the use of one-hot encoding in finite state machines, ECC protection of memories and periodical checks of the correctness of Zynq’s configuration memory. Two instances of the CIRIS data processing stages are used to increase the reliability of the system, simultaneously processing the data delivered by the same photo-detector and repeating the computation if there is a divergence in the results. The two processing stages can also work independently of each other to improve performance by processing data collected from different photo-detectors in parallel. Fault-diagnostic components are also included in the APEX-SoC to detect deviations from the expected instrument functioning, such as stuck-at situations in data transfers, malfunctions of the electromechanical components, die overheat situations or anomalies in the power supply. It is important to note that at this stage of the project we have not implemented any fault-tolerance mechanism on the processor yet, i.e. we rely on using the Cortex-A9 processor in the Zynq.
The CIRIS controller implemented on the APEX-SoC quadruples the performance requirement of the instrument as it is able to process spectroscopy data delivered by up to 100 photo-detectors (762 Mb/s), while the initial requirement was only 25. Besides, the APEX-SoC-based CIRIS controller only consumes around 5W, which is almost 40% less energy of that consumed by an equivalent board based controller. Finally, when irradiating the instrument photo-detector with a 60Co ɣ-ray source (1 rad/sec: approx. 3,400 hits/sec), the controller is able to compute a spectrum with the same spectral content of that corresponding to the non-irradiated spectroscopy data and with a comparable Signal to Noise Ratio (SNR).
Dr Iturbe spent the second and final year of the fellowship at the host institute, ARM, Cambridge in the UK. He focussed on a thorough study of the ARM CPU family to match the requirements of future space applications captured at NASA. Cortex-R5 CPU is identified as the most suitable ARM CPU, and investigated an architecture-level radiation tolerance technique. Dr Iturbe implemented a triplication model at the core level to build a radiation-tolerant Cortex-R5 CPU sub-system. The three Cortex-R5 CPUs are lockstepped sharing the caches and memories that are protected by SECDED code. He set up a fault injection methodology to emulate the radiation effects on the triple-core Cortex-R5 CPU sub-system. The sub-system is also implemented using commercial libraries to measure the performance, area and power efficiency overheads of the radiation-tolerant Cortex-R5 sub-system and compared to the non-radiation tolerant Cortex-R5.
Being a worldwide company with headquarters in Europe and leader of high-performance and energy-efficient mobile and embedded processor market, ARM has a big opportunity to become the leader of high-performance and energy-efficient radiation-tolerant processors. This will open the door to affordable space technology and the creation of a community of hardware and software developers (i.e. researchers and engineers) around it. In the long- term, the space/aerospace community could be integrated into the mainstream ARM eco-system, allowing for sharing the developer skills and expertise in the commercial mobile and embedded systems with the space community to create synergies around the ARM processor technology. This would have a tremendous impact on the European competitiveness.