A brief discussion on the origin and structural principle of power MOSFET

　　What is a Power MOSFET

　　We all know how to use diodes to make switches, however, we can only turn them on and off, not gradually control the flow of a signal. Furthermore, the diode as a switch depends on the direction of the signal flow; we cannot program it to pass or block a signal. For applications such as "flow control" or programmable switching, we need a three-terminal device and the bipolar triode. We have all heard of Bardeen & Brattain, who invented the triode by accident, like many other great discoveries.

　　Structurally, it is made of two junctions back to back (not a big deal, we probably used the same structure to achieve a common cathode long before Bardeen), but functionally it is a completely different device, like a "faucet" that controls the flow of emitter current - the "hand" operating the faucet is the base current. The bipolar transistor is therefore a current controlled device.

　　Field effect transistors (FETs) provide the same "faucet" function, although they are different in structure. The difference is that FETs are voltage controlled devices; you don't need base current, but use voltage to implement current control. The bipolar transistor was born in 1947, and soon after a brilliant father and son duo, Shockley and Pearson, invented (at least the concept) the FET. To distinguish it from its earlier bipolar "twin", the three electrodes of the FET are called drain, gate and source, and the three electrodes of the corresponding transistor are collector, base and emitter. There are two main variants of FET, which are optimized for different types of applications. JFET (junction FET) is used for small signal processing, while MOSFET (metal oxide semiconductor FET) is mainly used in linear or switching power supply applications.

　　Why was the power MOSFET invented?

　　When bipolar transistors are scaled up for power applications, they reveal some annoying limitations. Sure, you can still find them in washing machines, air conditioners and refrigerators, but these are low-power applications for us average consumers who can tolerate a certain level of inefficiency in home appliances. Bipolar transistors are still used in some UPS, motor control or welding robots, but their use is actually limited to applications below 10KHz and they are being phased out at an accelerated pace in cutting-edge technology applications where overall efficiency is a key parameter.

　　As a bipolar device, the transistor relies on minority carriers injected into the base to "beat" the recombination (electrons and holes) and be injected again into the collector. In order to maintain a large collector current, we need to inject current into the base from the emitter side, and if possible, recover all the current at the base/collector boundary (meaning that recombination at the base is kept to a minimum).

　　However, this means that when we want the transistor to turn on, there are a large number of minority carriers with a low recombination factor in the base, which the switch has to deal with before closing. In other words, the stored charge problem associated with all minority carrier devices limits the maximum operating speed. The main advantage of the FET now shines a light: as a majority carrier device, there is no stored minority charge problem, so its operating frequency is much higher. The switching delay characteristic of the MOSFET is entirely due to the charging and discharging of parasitic capacitances.

　　One might say: In high frequency applications, you need a fast switching MOSFET, but why would I want to use this device in my relatively slow circuit? The answer is straightforward: to improve efficiency. The device has both high current and high voltage during the duration of the switching state; because the device operates faster, less energy is lost. In many applications, this advantage alone is enough to compensate for the slightly higher conduction losses of the higher voltage MOSFET. For example, without it, a switch mode power supply (SMPS) with a frequency of more than 150KHz would not be possible at all.

　　Bipolar transistors are driven by current. In fact, because the gain (the ratio of collector and base current) decreases significantly with the increase of collector current (IC), the greater the current we want to drive, the greater the current we need to provide to the base. As a result, the bipolar transistor begins to consume a lot of control power, thereby reducing the efficiency of the entire circuit.

　　To make matters worse, this disadvantage is exacerbated at higher operating temperatures. Another consequence is the need for fairly complex base drive circuits that can pump and sink current quickly. In contrast, the (MOS)FET is a device that consumes virtually zero current at the gate; typical gate currents are less than 100nA even at 125°C. Once the parasitic capacitances are charged, the leakage current provided by the drive circuit is very low. In addition, the circuits for driving with voltage are simpler than those for driving with current, which is another reason why the (MOS)FET is so attractive to design engineers. On the other hand, its main advantage is the absence of secondary damage mechanisms. If you try to block large amounts of power with a bipolar transistor, the inevitable local defects in any semiconductor structure will act as current collectors, resulting in local heating of the silicon. Because the temperature coefficient of resistance is negative, the local defect will act as a low-resistance current path, causing more current to flow into it, increasing self-heating, and eventually irreversible destruction. In contrast, the MOSFET has a positive thermal coefficient of resistance.

　　On the other hand, the disadvantage of an increase in RDS(on) with increasing temperature can be felt, since this important parameter roughly doubles between 25°C and 125°C due to the reduction in carrier mobility. On the other hand, this same phenomenon brings a huge advantage: any defect that attempts to act as described above will actually shunt current away from it - we will see a "cooling spot" instead of the "hot spot" characteristic of a bipolar device! An equally important consequence of this self-cooling mechanism is the ease of paralleling MOSFETs to improve the current performance of a device.

　　Bipolar transistors are very sensitive to paralleling, and precautions must be taken to divide the current equally (emitter stabilizing resistors, fast response current sensing feedback loops). Otherwise, the device with the lowest saturation voltage will divert most of the current, resulting in the above-mentioned overheating and ultimately a short circuit.

　　Note that MOSFETs can be connected in parallel without any other precautions, other than designing safe symmetrical circuits and balancing the gates, so that they open equally, allowing the same amount of current to flow through all transistors. Moreover, the benefit is that if the gates are not balanced and the channels open to different degrees, this will still result in some drain currents being slightly larger than others under steady-state conditions.

　　对设计工程师有吸引力的一个有用功能是MOSFET具有独特的结构：在源极和漏极之间存在“寄生”体二极管。尽管它没有对快速开关或低导通损耗进行最优化，在电感负载开关应用中，它不需要增加额外的成本就起到了箝位二极管的作用。

　　MOSFET Structure

　　The basic idea of ​​the JFET (Figure 1) is to control the current flowing from source to drain by adjusting (pinching off) the cross-sectional area between the drain-source channel. This is achieved by using a reverse biased junction as the gate; its (reverse) voltage modulates the depletion region, which pinches off the channel and increases its resistance by reducing its cross-sectional area. Since there is no voltage applied to the gate, the resistance of the channel is at its lowest value and the drain current flowing through the device is at its highest. As the gate voltage increases, the beginnings of the two depletion regions advance, reducing the drain current by increasing the channel resistance, until total pinching occurs when the beginnings of the two depletion regions meet.

　　MOSFETs exploit the characteristics of MOS capacitors using different types of gate structures. By changing the value and polarity of the bias applied to the top electrode of the MOS structure, you can drive the chip below it all the way to inversion. Figure 2 shows a simplified structure of an N-channel MOSFET, known as a vertical, double-diffused structure, which starts with a highly concentrated n-type substrate to minimize the bulk resistance of the channel portion.

　　On top of it, a layer of n-epi is grown, and two continuous diffusion regions are made. Appropriate biasing in the p region will produce the channel, while the n+ region diffused into it defines the source. Next, after forming phosphorus-doped polysilicon, a thin high-quality gate oxide layer is grown to form the gate. Contact windows are opened in the top layer to define the source and gate electrodes, while the bottom layer across the wafer makes contact to the drain. Since there is no bias on the gate, the n+ source and n drain are separated by the p region, and no current flows (the transistor is off). If a positive bias is applied to the gate, the minority carriers (electrons) in the p region are attracted to the surface under the gate plate. As the bias voltage is increased, more and more electrons are confined in this small space, and the local "minority" concentration becomes greater than the hole (p) concentration, resulting in "inversion" (meaning that the material under the gate immediately changes from p-type to n-type). Now, an n "channel" has formed in the p-type material under the gate structure that connects the source to the drain; current can flow. Just as in a JFET (although the physics is different), the gate (by virtue of its voltage bias) controls the current flow between the source and drain.

　　There are many MOSFET manufacturers, and almost every manufacturer has its process optimization and trademark. IR pioneered HEXFET, Motorola built TMOS, Ixys made HiPerFET and MegaMOS, Siemens has the SIPMOS family of power transistors, and Advanced Power Technology has Power MOS IV technology, to name a few. Whether the process is called VMOS, TMOS or DMOS, it has a horizontal gate structure and the current flows vertically through the gate.

　　The special feature of power MOSFET is that it contains multiple "units" as described in parallel connection in Figure 2. MOSFETs with the same RDS(on) resistance are connected in parallel, and their equivalent resistance is 1/n of the RDS(on) of one MOSFET unit. The larger the die area, the lower its on-resistance, but at the same time, the larger the generated capacitance, so its switching performance is worse.

　　If everything is so strictly proportional and predictable, is there any way to improve it? Yes, the idea is to minimize (lower) the area of ​​the basic unit, so that more units can be integrated in the same footprint, so that RDS(on) decreases and the capacitance remains the same. In order to successfully improve each generation of MOSFET products, it is necessary to continuously make technological improvements and improve the wafer manufacturing process (better line etching, better controlled implants, etc.).

　　However, continuous efforts to develop better process technology are not the only way to improve MOSFETs; changes in conceptual design may greatly improve performance. Such a breakthrough is the successful development of TrenchMOS process announced by Philips last November. Its gate structure is not parallel to the surface of the die, but is now built in the channel, perpendicular to the surface, so it takes up less space and makes the flow of current truly vertical (see Figure 3). With the same RDS(on), Philips' transistor reduces the area by 50%; or, with the same current handling capacity, reduces the area by 35%.

　　Conclusion

　　We have compared the MOSFET to the more famous and more commonly used bipolar transistor, and we have seen the main advantages that the MOSFET has over the BJT, and we are now aware of some of the trade-offs. The most important conclusion is that the efficiency of the entire circuit is determined by the specific application; the engineer must carefully evaluate the balance of conduction and switching losses under all operating conditions and then decide whether to use a conventional bipolar, a MOSFET, or perhaps an IGBT?