| 摘要: | 台灣半導體居全球領先地位，高端的半導體已經達到5奈米的技術水準，半導體的封裝製程水準決定了半導體的品質。目前半導體的打金焊球凸塊(gold stud bump)製程主要是在半導體晶圓或是單顆積體電路(IC)上用金線焊成金焊球凸塊。以目前業界對於金焊球凸塊的製程水準要求，金焊球凸塊的厚度尺寸的誤差容許範圍大約只有正負5微米，但因為凸塊的厚度尺寸很小造成高度的量測困難且無法即時量測，目前都以控制焊線頭輸出的超音波電流與鍵合力量再配合品檢分析來調整製程的條件以維持良率，但此方法不夠精準，為業界普遍的製造瓶頸。
本論文實驗將振動感測器裝在焊線頭主軸上，再以打線機的最佳參數進行微調，再以每秒收集3次的頻率，全程即時量測在改變超音波電流時對應的振動數據與10秒計算出一次的機械衰減率(Decay rate)，再配合良率的結果並拍照比較，每組實驗都會在金焊球凸塊焊接成功的情況下去量測金焊球凸塊的截面積尺寸與厚度。本實驗會做5組數據並將收集的數據經過學習分析建立資料庫，藉以調控最佳的生產條件。
由照片結果顯示，只改變最佳參數的超音波電流後，金焊球凸塊的橢圓截面積的長短軸尺寸的範圍在59??m~83??m與凸塊厚度高度落在13??m~20??m範圍在時，振動感測器每分鐘的平均振動加速度會落在2.285m/s2 ~ 2.444m/s2之間,且生產過程中的機械平均衰減率(Decay rate)介於409~610之間。當超音波電流增加到120mA時，金焊球凸塊橢圓截面積的尺寸會超出規格落在86??m~91??m，此時振動感測器每分鐘的平均振動加速度會大於2.483m/s^2，而生產過程的機械平均衰減率達到611。當超音波電流降低到50mA時，金焊球凸塊的厚度尺寸會超出規格落在21??m~23??m，此時振動改測器每分鐘的平均振動加速度會小於2.284m/s^2，生產過程的機械平均衰減率更高到648。
本論文研究結果顯示，打線機超音波電流的大小變化會同步反映在焊線頭主軸的振動變化與機械衰減率然後也會直接影響金焊球凸塊的尺寸。結果顯示每分鐘平均振動值在2.285m/s^2 ~ 2.444m/s^2範圍內，此時衰減率也低於臨界值610以下，金焊球凸塊的截面積長短軸尺寸與厚度可以控制在合格尺寸，換句話說藉由即時偵測打線設備裡焊線頭的振動變化可得到99%以上的良率。
;Taiwan′s semiconductor industry holds a leading position globally, with high-end semiconductors reaching the 5nm technology level. The assembly quality of semiconductors determines their overall quality. Currently, in the gold stud bump process of semiconductors, the procedure involves using gold wire to form gold bumps on semiconductor wafers or individual integrated circuits (ICs). According to the current process requirement for gold stud bump quality, the allowable bump thickness deviation is only ±5 micrometers. However, due to the small size of the gold bumps, it is difficult to measure their height accurately and in real-time. At present, the process conditions are adjusted to maintain yield by controlling the ultrasonic energy and bonding force of the wire bonder and through quality inspection analysis. Nonetheless, this method is not precise enough, representing a common manufacturing bottleneck in the industry.
The experiment in this paper involves attaching a vibration sensor to the spindle of the wire bonding head. The optimal parameters of the wire bonding machine are then fine-tuned. Vibration data corresponding to changes in ultrasonic energy are collected in real-time at a frequency of three times per second, and the mechanical decay rate is calculated every ten seconds. The results are compared with yield rates and photographed. For each experiment, the cross-sectional size and thickness of the gold bump are measured under successful bonding conditions. This experiment will be conducted with five sets of data, which will be analyzed and used to build a database to adjust the optimal production conditions.
Results from the photographs indicate that when only the ultrasonic energy of the optimal parameters is changed, the range of the major and minor axes of the elliptical of the gold stud bump is between 59μm and 83μm, and the bump thickness falls within 13μm to 20μm. During this range, the average vibration intensity measured by the vibration sensor is between 2.285 m/s2 and 2.444 m/s2, and the average mechanical decay rate during production is between 409 and 610. When the ultrasonic energy is increased to 120mA, the major axis dimension of the gold stud bump′s elliptical cross-sectional exceeds the specification, ranging from 86μm to 91μm. At this time, the average vibration intensity per minute measured by the sensor exceeds 2.483 m/s2, and the average mechanical decay rate reaches 611. When the ultrasonic energy is decreased to 50mA, the thickness of the gold stud bump exceeds the specification, falling between 21μm and 23μm. At this time, the average vibration intensity per minute measured by the sensor is less than 2.284 m/s2, and the average mechanical decay rate increases to 648.
The results of this paper indicate that changes in the ultrasonic energy of the wire bonder are synchronously reflected in the vibration changes and mechanical decay rate of the spindle, which directly affect the dimensions of the gold stud bumps. The results show that when the average vibration value per minute is within the range of 2.285 m/s2 to 2.444 m/s2 and the decay rate is below the critical value of 610, the major and minor axis dimensions and the thickness of the gold stud bumps can be controlled within acceptable limits. In other words, by real-time monitoring of the vibration changes in the wire bonder′s spindle, a yield rate of over 99% can be achieved. |
