Paper Presentation

Paper Presentation SimTag: Exploiting Tag Bits Similarity to Improve the Reliability of the Data Caches Yun-Chung Yang Jesung Kim, Soontae Kim, Yebin Lee 2010 DATE(The Design, Automation, and Test in Europe)

Outline • Abstract • Introduction • Related Work • Proposed Method • Experiment Result • Conclusion

Abstract Though tag bits in the data caches are vulnerable to transient errors, few effort has been made to reduce their vulnerability. In this paper, we propose to exploit prevalent same tag bits to improve error protection capability of the tag bits in the data caches. When data are fetched from the main memory, it is checked if adjacent cache lines have the same tag bits as those of the data fetched. This similarity information is stored in the data caches as extra bits to be used later. When an error is detected in the tag bits, the similarity information is used to recover from the error in the tag bits. The proposed scheme has small area, energy, and performance overheads with error protection coverage of 97.9% on average. In contrast, the previously proposed In- Cache Replication scheme is shown to incur large performance and energy overheads.

Introduction • The idea of this paper come from the figure below. • Because of the transient error of tag bit • False-hit • Refer cache miss as cache hit. • False-misses • Make cache hit as cache miss. • Replacement errors • The above both cause the replacement error.

Related Work Protection SEC-DED(Single Error Correction Double Error Detection) EDC(Error Detection Code) ECC(Error Correction Code) A Framework for Correction of Multi-Bit Soft Errors in L2 Cache Based on Redundancy[12] Vulnerability analysis of L2 cache elements to single events upset[6] ICR: In-cache replication for enhancing data cache reliability[4] This paper

Proposed Method • The proposed architecture use four additional component to implement the SimTag. • Shifter • STI Encoder • STI Replace Handler • Error Corrector • Main Controller • STI(Same Tag Information) bits consist three logic parts • A valid bit • A set location bit • Way location bit

Reduced-Tag Cache Architecture

Components • Shifter • Use shifter to access lower and upper cache lines. • Another way – counter base • STI encoder • To generate STI bits, compared the tag bits of cache miss with tag bits from lower and upper sets.

Components(II) • STI Replacement Handler • Check STI bits on cache replacement, if the STI bits points to a replacement, it invalidate the STI valid bit and generate the new STI bits. • Error Corrector • Correct the tag bit when error detected by using STI bits. • Main Controller • If cache miss or tag bits errors, stalled the pipeline and signal the additional shifter or counter to access adjacent sets.

Experiment Result • Experiment Setup • Error coverage • Performance • Energy consumption

Experiment Setup • XEEMU, an improved Intel Xscale PXA80200 power simulator. • Implement the proposed method and ICR. • Use MiBench for experiment evaluation.

Error Coverage • ICR shows lower error coverage. • The average of our scheme is 97.9% while ICR covers 95.7 at most.

Performance • The performance of the ICR(in-cache replication), due to the dead block problem, I can suffer from significant performance.

Energy consumption • Addition energy consumption • Average energy consumption increases is 0.4%. • Datapath energy consumption • The proposed method does not increase the datapath energy consumption.

Energy consumption(cont.) • DRAM energy consumption • Due to the dead block of ICR(in-cache replication) in the main memory. This increase the DRAM energy consumption. While the proposed method does not use the main memory.

Conclusion • Most of the previous technique focus only data bits in the data cache. • Faulty tag bits can be simply replaced with correct tag bit from the adjacent cache line for error correction. • Our technique shows 97.9% error coverage with low additional energy consumption.

Paper Presentation