二維複合材料之介質常數與熱導特性分析研究

Abstract

長期以來高科技公司研發遵循著摩爾定律，電晶體密度的持續增加使得邏輯晶片的運行效果日益精進，而鰭式場效電晶體等三維堆疊架構之創新的元件設計則又讓積體電路的方向朝著3D IC更進一步發展。但高密度的電晶體也理所當然地在運作時產生了更高的熱量，若未有效散熱會使得其容易因為較高的溫度有危害。除此之外，現在的5G通訊與RF射頻電路的需求供不應求，不過這些高頻的元件往往會在電子傳送過程中面臨延遲或是能量損失的問題。想要讓上述的問題迎刃而解，開發具有高熱導且低介電值與介電損耗的封裝複合材料是刻不容緩的。 傳統上的封裝材料為聚丙烯(PP)之類的可塑材質，這類的材料通常都有熱導係數低（~0.2 W/mK）的缺點。一般來說，要提升其熱導係數研究人員多半會考慮就是參入導熱性高的陶瓷粉末。但是即便如此，混入陶瓷粉末的複合材料其熱導係數也只能提升至約1W/mK左右。相比起金屬的導熱性仍然相差甚遠，無法有效地將熱能導到外面。 我在第三章的地方用3ω與hot disk這兩種垂直熱導量測方法來測量pre-mixed graphene/CNF材料與環氧樹脂(BFE170)混成型的封裝材質。這種pre-mixed graphene/CNF材料混合了石墨烯 (graphene)、奈米纖維素(CNF)、環氧樹脂 (Epoxy)。透過他們的交錯分布互交織支撐，形成了緻密的散熱網路以增強其熱導能力。然而我們發現到剛混合好尚沒有固化完成之溶液中會有許多的顆粒產生，而這些顆粒導致了很困難製作3ω樣品。在選擇過濾顆粒的過濾器上我花了一些時間，最終選擇出了適合的過濾網並完成了改進後的3ω樣品。而兩種方法量測出來的複合材料之熱導值相近，因此得以驗證3ω與hot disk之垂直熱導量測結果。 而介質常數與介質損耗正切比的部分，則是承接了之前建良學長的研究對pre-mixed h-BN/CNF 做進一步的分析。而pre-mixed h-BN/CNF 的製做原理也與pre-mixed graphene/CNF類似，將六方硼氮(h-BN, Hexagonal boron nitride)與奈米纖維素(CNF)構成散熱網路的pre-mixed h-BN/CNF混合入聚丙烯(PP)中。值得注意的是，介質常數與介質損耗正切比(dielectric loss tangent)並不會與材料成分比完全正相關，要如何精準的預測其數值並不容易。以前學者多採取Bruggeman model模型來計算，但針對本研究的材料系統含有二維結構單元，其不一樣朝向之介質常數並不一致。從Bruggeman model模型中導入了 λ 修正項來修正混入二維材料的誤差。而介質損耗正切比(loss tangent)的部分則是基於Laplace transform model的理論，透過ψ 與 "ξ" 這兩個修正項來消除電磁波在材料中穿過h-BN時與因為頻率變化。

The semiconductor industry has been developing. The increasing density of transistors continues to improve the computing capability of logic chips. The innovative design of transistors, such as FinFET and GAA, has enabled the development of integrated circuit approaches to 3D ICs. However, the high density of transistors generates additional heat as devices operate, leading to higher temperatures that may damage the transistor itself. Additionally, 5th-generation mobile communication technology and RF ICs are all progressing into the high-frequency field. The signal from the high-frequency device is easily delayed or lost during transmission, which contributes to signal distortion. To solve the above dilemma, researching packaging composite materials has high thermal conductivity, and low dielectric constant and low dielectric loss tangent can bring fabulous market value. Traditionally, thermoplastic materials such as polypropylene (PP) have been adopted for the IC packaging industry; however, these types of materials have the disadvantage of low thermal conductivity (~ 0.2 W/mK). Generally, the most common way to increase thermal conductivity is to mix it with high-thermal-conductivity ceramic powders. Even so, the composite materials k with ceramic powder can only reach ~ 1.0 W/mK, which is remaining below than the k of metals. Obviously, it is not enough to effectively dissipate the amount of heat energy for 3D IC. In this thesis, we employ the 3ω method and hot disk approach to discover the k of a pre-mixed graphene/CNF material developed by our TFANL Lab. The pre-mixed graphene/CNF material consisted of graphene, Cellulose Nanofibers (CNF), and Epoxy. By interweaving one-dimensional (1D) and two-dimensional (2D) materials, a dense thermal dissipation network was formed to raise the composite k. Nevertheless, we found that the composite solution contained numerous particles before epoxy curing. The particles make the fabrication of the 3ω sample very hard. Therefore, we took some time to select the filter that would remove the particles from the solution. Finally, we use the proper filter to finish the 3ω sample. The k detected by the 3ω method and the hot disk approach are close, which verifies the two thermal measurement methods. As for the dielectric constant and loss tangent, I furthered Chine Liang’s work and research on the pre-mixed h-BN/CNF material. The principle of pre-mixed h-BN/CNF is similar to the pre-mixed graphene/CNF material. The h-BN and CNF construct an effective thermal dissipation network, and the pre-mixed h-BN/CNF material is combined with PP to enhance composite k. As for dielectric constant and dielectric loss tangent, as well as the proportion of material composition, is not linear, it is not easy to calculate the parameters precisely. Most traditional methods use Effective medium approximations (EMAs), such as the Bruggeman model, to calculate the properties. However, the composites we developed contain 2D materials, and their dielectric properties vary in different directions. Therefore, we introduced a λ correction coefficient into the Bruggeman model to correct the error. The dielectric loss tangent is based on the theory of the Laplace transform model, and the fluctuation due to electromagnetic waves passing through 2D material as frequency variation was corrected by two correction coefficients, ψ and "ξ".