MITIGATING RADIATION EFFECTS IN NETWORK ON CHIP LOGIC

This thesis investigates the impact of proton radiation on Networks-on-Chip (NoCs),
implemented in Field Programmable Gate Arrays (FPGAs), addressing a gap in
the existing literature on the vulnerability of network routers when exposed to the
radiative environment one encounters in space. We confirmed the vulnerability of
routers in a radiation test campaign, using a cyclotron, which provided the proton
radiation necessary to emulate space environmental conditions, by implementing
a NoC on a Basys3 FPGA to understand its fault characteristics.
In the context of this research, we investigate router vulnerabilities and propose
countermeasures to mitigate radiation-induced faults in the router. Addressing
NoC radiation failures is particular important for the future deployment of data
centers in space, which will require a tight integration of a multitude of process-
ing units (cores, accelerators, etc.) in a single chip, while operating in radiative
environments, and with limited shielding.
Several network topologies rely on routers to redirect traffic. For example, in
a star topology a single central router relays messages, forming a single point of
failure, and mesh topologies leverage one router per node to grant nodes access to
the network and to relay messages passing that node. While mesh networks can be
made robust against some failures in individual routers, a node’s router remains a
single point of failure for that node to communicate.
This thesis confirms the suscepbility of NoC routers to radiation failures, by
presenting the results of a radiation test conducted on a star-topology NoC, and
evaluates countermeasures to mitigate and recover from such failures. Redundancy
(e.g., duplicating the star topology’s central router) is one such well understood
solution, but we will highlight the limitations of this solution under particle storms,
as they can occur during solar flares. Specifically, redundancy alone will not be
sufficient on the long run, when routers contain dynamic state that radioactive
particles may change. For that reason, we will also investigate repair through
dynamic router reconfiguration which is less well explored in the literature.
We explore how high traffic throughput can be maintained in critical situa-
tions by building radiation-resistant NoCs that enable us to correct errors without
needing to reboot the entire system or, on FPGAs, re-load the entire configuration
bitstream.
Reconfiguration can happen in two ways: by preparing a router with and
reloading an internal configuration memory, which updates how the router relays
messages; or by partially reconfiguring the router logic, which mitigates router
failures as long as the FPGA fabric remains operational.
However, as their complexity evolves, along with the complexity of Multiproces-
sor Systems on Chip (MPSoCs) in which FPGAs are often integrated, it becomes
crucial to address their security more thoroughly when it comes to the NoCs implemented in FPGAs, which strongly impact the integrity and availability of the
design.
Although academia and industry have partially addressed these issues with
security units to prevent unauthorized entities from uploading bitstreams, error
correction to preserve bitstream integrity, and bitstream encryption, the solutions
tend to be board-specific, require additional manufacturing safeguards, or only
target simpler flaws such as bit inversions.
This thesis investigates a system capable of proactively mitigating radiation-
induced failures. However, since satellites and spacecraft are also increasingly
exposed to cyberattacks, some of which may well reach the lowest systems soft-
ware level, focusing on radiation-fault mitigating countermeasures is not enough,
in particular, if systems software triggers and controls these measures. For this
reason, this thesis also evaluates NoC radiation tolerance, subsequently analyze
the behavior of the solutions presented here on a large scale using a simulator