- 學位論文
用於矽光子平台之多模態波導交錯結構設計
Design of Multimode Waveguide Crossings for Silicon Photonics Platform
摘要
近日人們對大量數據處理的需求日增，電腦運算速度以及資料傳輸的速度因而要求不斷提升，對半導體及通訊產業形成很大的挑戰。其中一個解決方案是縮小電晶體的尺寸使同一塊晶片的電晶體數目增加以提升處理器的效率，但在晶片上的特徵線寬縮小到130 nm以下，傳輸電路的寄生電阻、電容所產生的訊號延遲大幅增加，使訊號傳輸率產生瓶頸。為了突破此瓶頸，有許多研究開始利用光連結取代電連結，將矽基光學元件整合到積體迴路中，在這之中低成本、低損耗的矽光子便成為大家積極研究的領域。 矽光子元件一般製作在SOI (Silicon On Insulator) 的晶元上，以單晶矽作為核心，二氧化矽作為包覆層材料，形成一個通道式或脊狀波導來傳輸光訊號。利用矽與二氧化矽的高折射率差異，可減少元件尺寸。而矽本身對於1310 nm和1550 nm (一般光學通訊傳輸波長) 這兩個波段的訊號都不會吸收，可以降低訊號傳遞的損耗。且跟傳統作為晶圓的三五族材料比起來，矽的成本較低，晶片的成本可以大幅降低。因此在電光整合的情況下，矽光子元件可以降低傳統電訊號傳輸的損耗及發熱問題。 使用波導來傳輸光訊號時，交錯結構可幫助設計者將元件的布局配置的更加緊緻，進而最大化單位面積晶片上的使用率。但普通的直波導交錯結構中，光訊號在交錯區域會產生大量的損耗及對交錯波導造成干擾，因此許多文獻針對只傳輸基礎模態的交錯結構做特殊設計，且也得到了很好的效率。但是近年來隨著資訊量的傳輸越來越大，模態多工 (Mode-Division Multiplexing) 技術便逐漸的興起，模態多工技術是在輸入端利用模態多工器將不同模態的光導入同一個寬度的波導，模態間的正交性質使訊號之間不會互相干擾，再利用分模解多工器將不同模態的光傳輸到光檢測器中轉成電訊號輸出。此方法跟單模傳輸比較起來可以增加傳輸的資訊量，然而也需要可傳輸多個模態的交錯結構來減少元件尺寸，因此本篇探討的是可以傳輸三個模態在模態多工系統所需的波導交錯結構，以滿足一般製程代工廠的限制為前提，在波長範圍1300 nm－1700 nm三個模態的傳輸損耗及串擾最小化的元件設計。 本論文利用脊狀波導的弱侷限性及多模干涉的自成像原理，搭配最佳化演算法設計出特殊的波導形狀，透過此流程所設計出的元件，在1300 nm－1700 nm三種模態的平均損耗為: TE_0=0.437 dB、TE_1=0.357 dB、TE_2=0.296 dB，三種模態的最大損耗為: TE_0<0.46 dB、TE_1<0.79 dB、TE_2<0.354 dB，最大串擾為: -29 dB，此頻寬在現今的論文中是非常大的且損耗跟其他可以傳輸三個模態的設計比起來是最低的，製程容忍度方面，本論文探討了四種可能的製程誤差情況，分別為：(1) 三階段蝕刻的寬度同時增減，(2) 一階段有變動寬度其他階段不變，(3) 結構可能因為製程時光罩片沒有對準造成元件前後偏移的情況，(4) 結構可能因為製程時光罩片沒有對準造成元件左右偏移的情況。分析結果顯示在 ±10 nm的製程誤差範圍內各個模態的傳輸損耗變動率不會超過 0.08 dB (最大2% 的穿透率降低)。此結構可用於模態多工系統中，使元件的布局更加緊緻以減少體積，且可以利用本篇的設計方法為基礎，再進一步的去設計出可以傳輸更多模態的交錯結構。
並列摘要
With the emerging demands for processing an increasingly large amount of data, the faster computing speed and higher data transmission rate are required. It becomes a big challenge to semiconductor and communication industries. One of the solutions is to reduce the dimensions of transistor and thus increase the number of transistors on the electronic integrated circuit (IC) chips so as to enhance the performance of processors used in the computational equipment. However, as the characteristic line width of an IC is reduced below 130 nm, the delay time which is caused by the parasitic resistance and capacitance increases significantly which leads to a bottleneck for signal transmission rate. Many researches utilize optical interconnect to replace electrical interconnect by integrating silicon-based optical components into electronics ICs. Therefore, silicon photonics, which is low loss and low cost, has become a promising research topic. Silicon photonic components are typically fabricated on an SOI (Silicon On Insulator) substrate with silicon as the core and silicon dioxide as the cladding to form channel or rib waveguides for transmitting optical signals. With the high refractive index contrast between silicon core and silicon dioxide cladding, the size of components can be reduced. The loss of light transmission in silicon waveguides is lower than other material because silicon does not absorb optical signals at 1310 nm and 1550 nm. Therefore, with such electro-optical integration using silicon photonic components, the high signal loss and heat generation problem of conventional electrical signal transmission can be avoided. When the optical signal is transmitted in a silicon waveguide, the waveguide crossings allow the photonic circuit designers to make circuit layout tight and thus maximize the utilization of chip real estate. However, in a simple waveguide crossing design, optical signal will diverge at the intersection region and signal in the cross-port will be interfered. Several particular designs of waveguide crossings for transmitting the fundamental mode have been reported and have achieved great performance. In recent years, as the demand for higher data transmission rate increases, the mode-division multiplexing (MDM) technology, which utilizes orthogonal guiding modes to transmit independent signals, is emerging. At the input side of a typical MDM system, the optical signals are multiplexed into distinct waveguide modes through a mode-division multiplexer and then the signals propagates in the bus waveguide which is wide enough to support the required number of modes. At the output side of the MDM system, the optical signals will be separated though a mode-division de-multiplexer and then directed to corresponding photodetectors to be converted in the form of electrical signals. Compared to single-mode communication systems, MDM systems can considerably increase the data communication capacity. In MDM systems, waveguide crossings with low crosstalk and small footprint are essential. Therefore, low-loss and low-crosstalk multimode waveguide crossings which allow to transmit three modes operated in the wavelength range of 1300 nm－1700 nm and comply with the general foundry limitations are discussed in this thesis. By using the weak confinement of rib waveguides and self-imaging principle of multimode interference, waveguide crossings with particular profiles are designed with the optimization algorithm. In the wavelength range of 1300 nm－1700 nm, the performances of the waveguide crossing designed through this process are: the average insertion loss TE_0=0.437 dB,TE_1=0.357 dB,E_2=0.296 dB and the maximum loss TE_0<0.46 dB, TE_1<0.79 dB,TE_2<0.354 dB and the maximum crosstalk is -29 dB. The insertion loss of the proposed device is lower than the waveguide crossing designs for three-mode MDM systems presented in the previous literature and its bandwidth is relatively large. As for the fabrication tolerance, four cases are discussed: (1) the simultaneous variation of widths of the structures for all the three-step etching process, (2) the variation of the widths of the structures for each one of the three-step etching process, (3) the displacement along the propagation direction of the structures for each one of the three-step etching process, and (4) the displacement perpendicular to the propagation direction of the structures for each one of the three-step etching process. The results of fabrication error analysis indicate that the insertion loss variation for each mode does not exceed 0.08 dB under the condition of ±10 nm variation for the four cases. The proposed structure can be adapted to increase the integration density for the photonic circuits utilizing mode multiplexing. Furthermore, this method can be applied to design waveguide crossings for MDM systems with an even larger number of modes.