線上影音
Home > ANSYS Q3D 教學 > Q3D Extractor Tips
The article is intended to introduce some very useful skills, and demonstrates on Q3D Extractor 2014 platform.
-
Quickly add many source/sink ports -- method I for fixed pitch
-
Quickly add many source/sink ports -- method II for objects on the same axis
-
問題與討論
-
Reference
1.1 What is "local reference"
When computing capacitance, by default, Q3D Extractor uses the voltage=0V at infinity as a reference. That is, there is charge at infinity. However, in some case, you may want to use a grounded net in the model (in the solution region) as a reference, that is "local reference". This will cause the potential at infinity is not fixed at 0V, and it depends on voltages inside the model, that is "float at infinity". [Float at Infinity] implies the "total charge" at infinity is zero.
1.2 How to set "Local Reference"
1.2.1 [Q3D Extractor] \ [Reduce Matrix] \ [Float at Infinity], and this is post-process
1.2.2 Select a object in model to set ground net (as reference).
Method I:[Assign Net] \ [Ground]
Method II:[Reduce Matrix] \ [Ground Net]
Please note that:the sequence of setting [floating at infinity] and [Ground net] will affect the C matrix result. Refer to here
1.2.3 Check the result by selecting reduce matrix
沒有設定float at infinity時,看任一行的數值總和不為0,0.64464+0.64471=1.28935¹1.3926。這表示在 求解區域內的object/model,與無窮遠處有電力線/電荷的交互作用存在 (無窮遠處存在參考地)。
設定float at infinity後,看任一行的數值總和為0 (電荷守恆),0.69245+0.69245=1.3849。這表示在 求解區域內的object/model,與無窮遠處沒有電力線/電荷的交互作用存在 (無窮遠處沒有參考地)。
C matrix result table對角線單元為自容Cself,其值為正,非對角線單元為互容Cmutual,其值為負。HELP內有說明,正負表示相對感應的關係,that is |Cself| = |sum of all Cmutual relative to the conductor|
讀者可以建一個只有單一導體的design,設定floating at infinity,看看此時的Cself是多少?
Ans:For single conductor, Cself is always 0 under floating at infinity.
電容必定是兩導體間的事C=epsilon*d/A,若不考慮無窮遠處有reference,那單一導體不會有自容值
2.1 General Manual Mesh Setting
[Modeler] \ [Model Analysis] \ [Analysis Object], to find small edge or sliver face to healing
- [Q3D Extractor] \ [Mesh Operations] \ [Assign] \ [Model Resolution]
通常是很密的結構DC solver mesh出現mesh leak error時採用的招數
- [Modeler] \ [Units]
通常是mesh遇到error message,提醒你更換小單位表示時才會用到
2.2 For Project with Curved Surface
[Q3D Extractor] \ [Mesh Operations] \ [Initial Mesh Settings]
這對於probe card、測試板廠,要模擬上百根20um圓形直徑與間距的探針,幫助pass initial mesh很常用的設定 。但其實更建議的做法是,在一開始建立3D model時,就以多邊形(faced by 8~12 segments)來表示圓形結構
2.3 For Project with Super Thin Layer (<0.1um)
Change units from [mm] to [mils]
Set [Surface Approximation]
Q3D v11(含)以前的版本,對於大面積且厚度<=0.1um的object求解,建議放大厚度求解,否則mesh會很辛苦,並且要換算等效介電係數與導電率。Q3D v12開始,可以支持導體厚度=0的求解 (Assign "Thin Conductor" boundary),這對於In-cell的應用,或是求解um等級的細線耦合量(不能放大厚度),非常有用
2.4 For Layered Structure as PCB or Projected Capacitive Touch Panel (常用)
[Q3D Extractor] \ [Mesh Operations] \ [Initial Mesh Settings]
[Class mesh] is always better for layer structure application
3.1 HPC (High Performance Computing)
- [Tools] \ [Edit Active Analysis Configuration]
3.1.1 Tasks指的是可以分散的工作數目,比方CG\DC_RL\AC_RL solve process如果都需要計算,就可以分三個Tasks並行運算
task不是越多越好,以上圖設定為例,每個task分到4 core。一般建議每個task分到6~8 core的前提下,task越多效能越好
3.1.2 Cores指的是核心數,不管是真正硬體核心數,或軟體加倍後的threads數目。(一般是建議關掉軟體threads)
3.1.3 可以使用IP Address或DNS Name指定網路內不同的machine來一起運算。
3.1.4 添加machine list。在互聯時,要關掉Window防火牆,且每台電腦各自都要裝Q3D與RMS
3.1.5 可以測試對machine的網路連線是否正常
3.1.6 指定要分散處理(並行處理)的工作(task)種類
3.1.8 如果在step7有選超過兩類的distribute type,而在step2設的task數目也夠多,那這裡就可以設two level(巢狀式分散處理)
- [Q3D Extractor] \ [Analysis Setup]
Q3D的solution frequency不會影響initial mesh,但會影響AC_RL matrix的計算結果,所以solution frequency必須設在AC region。
Q3D只有DSO沒有DDM
3.2 Make sure the number of mesh enough -- set [Min. Converged Passes]=2用來避免假收斂
3.3 Speed up convergence for complex structure
舊版的Q3D solution order分三檔:Normal, High, Very High;自Q3D2015.1(R16.1)起,solution order分四檔:Normal, High, Higher, Highest
3.3.1 For case1(5x5 patterns, 2 layer ITO) with real thickness 10um and Q3D2015.1
Normal order, percent error 0.5% -- it takes 8:12, more mesh but much faster
Highest order, percent error 0.5% -- it takes 17:21, much slower
3.3.2 For case1(5x5 patterns, 2 layer ITO) with thin conductor 0.1um and Q3D2015.1
Normal order, percent error 0.5% -- it takes 4:26, more mesh but much faster
Highest order, percent error 0.5% -- it takes 23:52, much slower
3.3.3 For case2(1x2 patterns, 1 layer ITO with VCOM) with real thickness 10um and Q3D2015.1
Normal order -- it takes 4:53, slower
Highest order -- it takes 2:36, faster
In general, set Solution Order as [Normal] by default is good (an001 3.3.1, an001 3.3.2), but for some cases that is hard to converge as an001 3.3.3, 6.1, 7.8, set Solution Order as [Higher\Highest] is better.
High order的作用是透過較高階數的矩陣數值運算,以較少的mesh而可以較快達到所要求的percent error。所以high order會讓mesh時間減少,但matrix solver時間卻增加。對於net數較少,或很難收斂的題目(percent error£0.1),建議可設較高階的order.
同一個題目,有可能你求解要求是0.5%時用[Normal]較快(an001 3.3.2),但若求解目標改成0.1%收斂兩次時,則[Highest]較快(Basic PCPT 7.8)。表示該題目以[Normal]很難收斂到0.1%以下
[Q3D Extractor] \ [Design Settings]
5.1 Create a sheet to set source/sink
5.2 [Edit] \ [Duplicate] \ [Along Line], then set duplicate number
5.3 Assign source/sink:[Q3D Extractor] \ [Nets] \ [Assign 2D Terminals]
6.1 Create a coordinate to snap the face of objects which need ports
[Modeler] \ [Coordinate System] \ [Create] \ [Relative CS] \ [Offset]
6.2 Change to [XZ] plane, then right-click to select [Faces On Plane]
6.3 Create sheet:[Modeler] \ [Surface] \ [Create Object From Face]
6.4 Assign source/sink:[Q3D Extractor] \ [Nets] \ [Assign 2D Terminals]
- 問題與討論
7.1 何時需要設定local reference?
Ans:筆者的習慣是,對於要計算電容的Q3D例子,都默認增加設定[Float at Infinity],但不另設[Ground Net]
7.2 為何只計算電容則不需要設source/sink?
Ans:計算C與計算R、L不同,不需要真的有電流流過導體就能求得C,故不需設source/sink
從R16版本起,Q3D求解DC_RL如果沒有設sink\source,validation check就能幫你檢查出來,提醒你注意喔
7.3 SI2D的solution type:[Open]、[Close]分別的使用時機?
Ans:只有coaxial cable的分析會用到[Close] type,與[surface ground] conductor.
7.4 在step1.2.2中,兩種方法設定local reference (ground)有沒有差異?
Ans:Method I與Method II的CG結果會有些微不同(mesh的數目就不同),但這差異不大(約0.6%)。如果要讓結果相同,把Percent Error設的更小一點(0.3%改0.1%),讓mesh切得更細即可。
直接設Grounded Net type或靠Reduce Matrix設Ground Net,CG與DC_RL結果一樣,但Ground Net RM with sink\source則會影響AC_RL。
以下是摘錄自HELP的說明:
1. "assign ground net" treats the conductor V=0, and affects C matrix only.
Grounded Net type不影響DC_RL與AC_RL,對C matrix的影響是把該conductor視為V=0並將其在C matrix隱藏起來。而對C matrix計算來說,把物件ground的另一方法是:指定導體材料, 但不對它做auto. identify net。
2. "ground net" reduce matrix with sink\source:nearby net's magnetic filed will induce current on the grounded plane surface, so it affects AC_RL matrix.
Matrix reduction is the only way to model grounded conductors in inductance and resistance computations.
只有使用Method II,才能在ground net上設定sink\source,也唯有如此才能考慮ground plane上的電流動方向,其影響AC_RL,不影響CG與DC_RL。有定義current loop所得到的AC region RL值才準確,若設一個沒有sink\source的ground net,其AC_R會偏小,AC_L會偏大,即只能表示低頻的特性 。
在Reduce Matrix設ground net或return path,兩者也不同。後者是用loop L=L1+L2-2*L12這種partial L的關係計算,會同時影響DC_RL與AC_RL。其適用於return current只在某一導體的封閉迴路內流動的情況,算出的AC_RL值會 比ground net RM得到的值略大。
Method I與Method II,若後者不在ground net上設sink\source,則兩方法結果相同。若後者在ground net上設sink\source,唯有這種做法才能表現AC_RL高頻的真實特性,因為有考慮induced current在ground plane上的流動。
設成ground net的金屬平面上流動的是周圍訊號線對V=0平面的感應電流,設成return path的金屬平面流動的則是所有訊號的電流(DC+AC)被強制迴流。
-
Reference