Microprocessors depend on high integrity instructions and
data to function properly. Various factors can corrupt this information, with
high-energy sub-atomic particles and other types of electromagnetic radiation
effects being among the more difficult of these factors against which to guard.
Because a single erroneous bit of information can lead to significant operation
errors, processors may employ error detection and correction (EDAC) techniques
to verify the integrity of information prior to use. Indeed, EDAC schemes can
even provide an effective safety-guard against radiation effects (radiation
hardening).
Still, existing EDAC schemes, while effective, produce
performance penalties (e.g. processing, area) that limit the use of EDAC schemes
to larger memories and second layer (L2) cache of high performance processor
applications. These performance penalties generally preclude the use of EDAC
schemes in register files and limit the application of EDAC schemes in first
level (L1) caches. Consequently, many processors must rely on error detection
schemes alone, reserving employment of EDAC schemes for memories where processor
overhead, access speeds, and overall size have limited importance.
Accordingly, there is a need for an EDAC scheme that is
significantly more efficient than conventional EDAC schemes. Moreover, for
environments subject to radiation, there is also a need for an efficient and
effective EDAC scheme that is conducive to radiation hardening.
Researchers at Arizona State University have developed a
novel lightweight EDAC scheme that employs a two-dimensional parity technique to
reduce area and timing overhead penalties. The technique is appropriate for high
speed memory structures such as register files and L1 caches and provides
radiation hardening by design.
Potential Applications
- Integrated Circuits (e.g. Microprocessors, etc.)
- High Speed Memory Structures (e.g. Register Files,
Caches, etc.)
- Radiation Hardening (e.g. Satellites, Space Exploration,
Defense, etc.)
Benefits and Advantages
- Provides Lightweight Error Detection and Correction –
- Reduces area and timing overhead penalties of
conventional EDAC schemes through the simplicity of the parity scheme; area
overhead varies between 2% and 12.7%, compared with 4% to 62.5%, depending
on code word row width
- Suitable for small, fast arrays such as register files
and L1 caches that require low read and write latency as well as small write
granularity
- Provides Radiation Hardening by Design
- Mitigated 100% of soft-errors using appropriate bit
interleaving during accelerated heavy ion testing
- Allows fabrication of hardened circuits from
commercially available state-of- the-art CMOS manufacturing
processes
Download original PDF