EEPW论坛

半导体的性能会随着温度的改变而改变。

　　在20世纪80年代早期，当三大汽车公司——通用，福特，克莱斯勒开始在每个车模中大大增加电子器件的时候，我已经断断续续的在大半导体公司做产品工程师了。特别当引擎温度问题和可靠性成为主要考虑时，这直接导致公司多数产品0 to 25°C的规格改为–40 to +125°C。消费者在抽样检查中拒绝三个有2N3906晶体管的货物出货，认为它们漏电流偏大。我接到一些“有缺陷”但没错误的样品。这个问题摆在了我的面前，在汽车生产厂商不得不关闭整个生产线之前，只有一周时间去解决。第二天，我就带着那些返回的部件以及一些相关的部件冲出去了。在我到达前，结构工程师告诉我说，当部件被加热到125°C时，所有的部件都没有通过30V、50-nA的漏电流测试。当我意识到他将室温规格的产品用到整个温度域的时候，我不得不闭上了的嘴。

　　这个工程师没有意识到半导体性能会随温度改变而改变。我拿出了从Intel奠基人Andy Grove得到《Physics and Technology of Semiconductor Devices》1967复印版，说明了漏电流是怎样随温 
度而变化的（参考文献1）。将那些公式和图表简化为一个基本规则：温度每上升10°C漏电流就会翻倍。这个解释不足于解决产品问题，所以我们在检验室用Tektronix 576绘图工具和TO-92加热探针测试相关的部件和产品抽样。所有部件都显示了期望硅元素的随温度改变的相同漏电流特性。

　　在会议后，结构工程师同意在产品大范围抽样测试后均可出货。我们将2N3906在25°C时50 nA的限制扩展为125°C时50 µA，并开始了抽样检查。

　　在最后的测试中，加热探针失效而不能用于后续的抽样测试。工程师又一次拒绝批准小范围抽样和晶体物理特性的部件。为了不出现任何导致三大厂商生产下跌的情形，我不得不提出一个解决方案。

　　再次掏出教科书，我证明了与死区温度的直接联系基射极电压。同样的，当用到硅元素时，又可以将公式简化为：每摄氏度减小饱和基射极电压2mV，即每增加100°C会减少200mV电压。使用Tek 576绘图器，用打火机加热部件时，可以观察到电压降。一旦部件到达或超出温度，改变漏电流模式，并确保部件完好。我使用这个方法完成抽样，结构工程师认可了。转天，产品经理告诉我，因为这周已经发出的汽车产品没配，他不得不每辆车配五个2N3906给经销商。代理商们不得不安装这些鉴定过的结构部件。当你的车门或是后备箱微开时，这些部件将驱动LED发光。

　　参考文献：

　　1．Grove, Andrew S, Physics and Technology of Semiconductor Devices, John Wiley and Sons, 1967.

　　英文原文：

　　Solid-state physics convinces customers

　　Tales From The Cube: Yes, semiconductor performance does change with temperature.

　　By Martin Delateur, Consultant -- EDN, 8/16/2007

　　I was working as a product engineer in the discrete division of a major semiconductor company in the early 1980s when the Big Three automotive companies—General Motors, Ford, and Chrysler—started to add many electronic accessories each model year. It was also when under-the-hood-temperature issues and reliability were major concerns, causing companies to change many 0 to 25°C specs to –40 to +125°C. Our customer rejected three shipments of 2N3906 transistors at incoming inspection, saying the parts had high leakage currents. I received some samples of the “defective” units but found nothing wrong. The problem reached my desk with only a week to resolve it before the automaker would have to shut down the production line. I flew out the next day with the returned units as well as some correlation units. Upon my arrival, the component engineer advised me that all the units were failing the 50-nA leakage test at 30V when he heated the units to 125°C. I had to close my gaping mouth when I realized he was applying the room-temperature specification across the full temperature range.

The engineer did n 
ot acknowledge that semiconductor performance should change with temperature. I pulled out my 1967 copy of Physics and Technology of Semiconductor Devices by Intel founder Andy Grove to show him how the leakage current increases with temperature (Reference 1). You can simplify the equations and graphs to the basic rule that leakage doubles for every 10°C increase in temperature. This explanation was not enough to release the product, so we went to the incoming-inspection lab to test my correlation units and production samples using a Tektronix 576 Curve Tracer and a TO-92 heating probe. All the parts showed the same leakage characteristics with temperature that you would expect from silicon.

　　After another meeting, the component engineer agreed to approve the shipments only after the lots had passed a large sampling plan. We expanded the 2N3906 limit of 50 nA at 25°C to 50 µA at 125°C and began to sample the lots.

　　During the final tests, the thermal heat probe failed and could not be used on the remaining lots. Once again, the engineer refused to approve the units based on a smaller sample and my solid-state-physics lessons. Not wanting to be any part of a department that caused a lines-down situation for one of the Big Three, I had to come up with a solution.

　　Again pulling out the textbook, I demonstrated how base-emitter voltage relates directly to the die temperature, as well. Again, you can simplify the equation when applying it to silicon, which reduces the saturated base-emitter voltage by 2 mV for every degree Celsius and translates to a 200-mV reduction in voltage for the added 100°C in temperature. Using the Tek 576 Curve Tracer, we could watch the voltage drop while heating the unit with my lighter. Once the unit met or exceeded temperature, we could change to leakage mode and confirm that the parts were good. I completed the sampling using this method, and the component engineer approved the parts. The next day, the production manager told me that he had to send out five 2N3906s per car to dealers, because that week's production of cars had been shipped without them. Mechanics at the dealership had to install the critical components, which drive the LEDs that light up when your door or trunk is ajar.