隨著製程的進步，串音(crosstalk)的影響越來越嚴重。在匯流排(bus)上，相鄰線路之間的串音問題會造成時間上的延遲(delay)；若能完全消除4﹒C串音的話，耦合電容(cross-coupling capacitance)所造成的延遲甚至能夠減少多達22%。因此，許多學者致力於消除串音的研究，也有許多論文已經被提出來。

在現有的消除串音的方法當中，多數是以插入保護線(shielding wire)、將線路重新排序(reorder)或者將資料編碼(encode)再傳輸…等方式來解決。這些方法都會產生許許多多的額外的硬體開銷，因此，我們希望能找出一個方法不需要額外的硬體開銷就能夠消除4﹒C串音。我們發現，指令匯流排上面所傳輸的資料順序是可以預先得知的，因此我們提出一個概念，也就是運用編譯器的技巧，將執行檔預先處理成為沒有4﹒C串音之執行檔，再將之送上指令匯流排去傳輸。

消除串音的方法包括下列三種：
1. 指令重排(Instruction reordering)：在不違背資料相依性的前提下，將指令重新排序。
2. 暫存器更名(Register renaming)：將可用的暫存器重新命名。
3. 插入無用指令(NOP insertion)：使用前兩個方法都無法消除的串音，只能靠插入無用指令來消除。這個方法是插入不影響程式機能且不會產生串音的指令，因此，會產生額外的指令開銷。

根據我們的實驗結果，僅需要2.14%到8.74額外的靜態指令，以及0.03%到9.73%額外的動態指令，就可以完全消除指令匯流排上的4﹒C串音，並可改善績效至少16.09%，最多達23.40%。

Crosstalk noise has become more and more important with the shrinking of technologies. Crosstalk between adjacent neighboring wires on the bus leads to problems of delay variation. Delay due to cross-coupling capacitance could be reduced about 22% when 4¢C crosstalk is eliminated.

Existing work on eliminating crosstalk, e.g. shield insertion and bus encoding, induces a lot of hardware overhead. Since data transmitted on instruction bus is known in advance, we propose a crosstalk elimination algorithm using compiler techniques, e.g. instruction reordering, register renaming and NOP insertion to
preprocess the program before it is transmitted on instruction bus and hence no hardware overhead is induced. Our experimental results show that with only 2.14% to 8.74% static instruction count overhead and 0.03% to 9.73% dynamic instruction count overhead, we can eliminate 4¢C crosstalk on instruction bus and improve performance by 16.09% to 23.40% as compared to the original program with 4¢C crosstalk.

[1] Bret Victor, Kurt Keutzer. Bus Encoding to Prevent Crosstalk Delay. 57-63 ICCAD, 2001.
[2] Chunjie Duan, Anup Tirumala, Sunil P. Khatri Analysis and Avoidance of Cross-talk in On-Chip Buses. 133-138 Hot Interconnects, 2001.
[3] Chunjie Duan, Sunil P. Khatri. Exploiting Crosstalk to Speed up On-chip Buses. 778-783 DATE, 2004.
[4] E. L. Lawler and D. E. Wood. Branch and bound methods: a survey. 699, 719 Operation Research, 14, 1966.
[5] Harmander S. Deogun, Rajeev R. Rao, Dennis Sylvester, David Blaauw. Leakage-and Crosstalk Aware Bus Encoding for Total Power Reduction. 779-782 DAC, 2004.
[6] Lei He, Kevin M. Lepak. Simultaneous Shield Insertion and Net Ordering for Capactive and Inductive Coupling Minimization. 56-61 ISPD, 2000.
[7] Peter Petrov, Alex Orailoglu. Compiler-Based Register Name Adjustment for Low-Power Embedded Processors. 523-527 ICCAD, 2003.
[8] P. Sotiriadis, A. Chandrakasan. Reducing Bus Delay in Submicron Technology Using Coding. 109-114 ASP-DAC, 2001.
[9] Siu-Kei Wong, Chi-Ying Tsui. Re-Con‾gurable Bus Encoding Scheme for Reducing Power Consumption of the Cross Coupling Capacitance for Deep Sub-micron Instruction Bus. 130-135 DATE, 2004.
[10] V. Zivojnovic, J.M. Velarde, C. SchlÄager, H. Meyr. DSPstone - A DSP-oriented Benchmarking Methodology. Proc. of the ICSPAT, 1994.
[11] Wu-An Kuo, Yi-Ling Chiang, TingTing Hwang, Allen C.-H. Wu. Performance-Driven Crosstalk Elimination at Compiler Level. Technical Report 05-01, Computer Science, National Tsing Hua University, 2005.
[12] Yan Zhang, John Lach, Kevin Skadron, Mircea R. Stan. Odd/Even Bus Invert with Two-Phase Transfer for Buses with Coupling. 80-83 ISLPED, 2002.
[13] http://www.spec.org/osg/cpu2000/CINT2000/. Spec2000 CINT Benchmarks.
[14] http://www.spec.org/osg/cpu2000/CFP2000/. Spec2000 CFP Benchmarks.
[15] http://www.simplescalar.com/.