The challenges facing the development of silicon carbide Schottky diodes for power applications are discussed by Roland Rupp and colleagues from Infineon Technologies.

SiC technology and device development are moving forward rapidly. Impressive results with SiC devices are regularly reported in this magazine; see for example C Weitzl et al. s article on the material s high-frequency applications (Compound Semiconductor May 2000, p45). On the other hand, total sales of SiC devices worldwide are to the best of our knowledge still close to zero with no significant upward momentum. The reasons for this are threefold. First, most of the existing devices are aimed at cost-insensitive niche markets like military applications or solar-blind UV detectors where the performance advantage is the key. Second, ultrahigh voltage applications, where high current ratings are required, cannot be addressed with SiC yet due to its high substrate defect density. Finally, semi-insulating substrates are needed for the high-frequency applications where SiC is truly the enabling technology. These are very expensive even compared with standard SiC substrates. Due to the low market volume of SiC devices, all manufacturing is done in a "laboratory production" style, and cost-effective production lines have not been built. To relieve this deadlock, we tried to determine the necessary preconditions for a mass market for SiC power devices. These are as follows:

The device should be as simple as possible to reduce the barrier to industrial volume production.

It should minimize the processing steps specific to SiC.

The required chip area (i.e. current rating) has to be low due to yield-limiting micropipe defects in the substrate.

High application competence is required to be able to convince potential customers that they can save money despite the high cost of the SiC device compared with a competing Si device.

The target market really has to be a mass market in order to allow a return of the development expenses.

The device we identified fulfilling all these requirements is the SiC Schottky diode for voltage levels of 300 and 600 V. Market forces System miniaturization is a trend in electronics that is mainly but not only driven by the increasing amount of portable applications. The power supply of these items is still a dominating part concerning the dimensions and weight of the whole system. For example, in a typical portable computer the power supply accounts for more than 10% of the systems total weight. Therefore all manufacturers of switch-mode power supplies (SMPSs) have defined roadmaps to increase the power density of their products. There are two major approaches to realizing these roadmaps. The first is to reduce the size of passive components by increasing the switching frequency at constant efficiency and/or reduce the size of the electromagnetic interference EMI filter through low noise generation. The second is to reduce power losses and correspondingly the cooling effort, reducing the need for a heat sink and fan. This results in a specific requirement for the main power semiconductor components: a significant reduction of switching power losses. This is why unipolar semiconductors such as MOSFETs and Schottky barrier diodes are replacing bipolar devices. The beauty of unipolarity is the absence of stored charge carriers and, therefore, theoretically instantaneous switching transients limited only by small parasitic capacitances. Power MOSFETs like CoolMOS and OptiMOS can be found in a wide range of blocking voltages from 20 V up to 1000 V. On the other hand, Schottky barrier diodes are restricted to 250 V, due to limitations of the silicon or GaAs material from which they are fabricated. There are several reasons for this limitation. One is very high leakage currents especially at higher temperatures, leading to reverse losses that are of comparable size to the forward losses. Another reason is the strong increase in the area-specific on-resistance (Ron, A) with Vb2.5, where Vb is the blocking voltage. Unfortunately this problem cannot be solved by simply increasing the chip area since this also increases the reverse losses. The input stage of an off-line SMPS usually requires devices rated in the range 500600 V. Thus a strong and unsatisfied need exists for Schottky barrier diodes capable of working in this voltage range. Cost-effective SiC Schottky diodes As mentioned above, a SiC Schottky diode is the SiC device with the simplest technological process for manufacturing. shows a schematic of the device layout. The whole device can easily be produced with a 34 mask layer process. There are three major functional items in this device: Drift region The epitaxially grown drift region is responsible for the maximum blocking voltage and the on-state resistivity. A well-controlled and homogenous epilayer growth is a precondition not only for SiC Schottky diodes. Within the Siemens/Infineon group we have a long history in developing epitaxial processes for SiC. The tool we are using today is a multiwafer reactor for six 2 inch wafers (Rupp et al.), while another reactor suited for seven 2 inch or five 3 inch wafers is in preparation (Vrob ev et al.). To adjust the trade-off between forward and reverse characteristics, the thickness and doping tolerance in the drift region have to be kept small. Our diodes are dimensioned for an average electrical field of 1.75 MV/cm. So it is clear that deviations in the +10% range for doping and 10% for thickness are already relevant for the reverse yield. Deviations in the other direction are conflicting with the forward performance; we have to be very stringent here as well, because we have to keep the area of the SiC devices small due to the still extremely high costs of the base material and the large intrinsic defect density (micropipes) in the order of 20/cm2. Guard ring The edge terminating guard ring determines the maximum blocking voltage at a given drift layer doping and thickness. The guard ring is generated by an acceptor ion implantation with subsequent annealing. Exact control of the annealing conditions is required to achieve the right concentration of electrically activated acceptors in the implanted area. Deviations in either the positive or negative directions lead to a reduced blocking capability (Mitlehner et al.). The semiconductor-metal junction This is responsible for the rectifying barrier and the leakage current of the diode in reverse direction. The biggest challenge is probably to get a reproducible and homogenous Schottky barrier between the SiC and the metal contact. Any kind of surface inhomogeneities usually lead to local barrier reduction and leakage current "hot spots". These can never be prevented completely. Sources of such inhomogeneities are surface impurities, locally differing reconstruction of the SiC surface and an inhomogeneous structure of the Schottky metal (i.e. epitaxial areas next to nearly amorphous regions). The effect of local barrier reduction on the forward characteristics of a Schottky diode is shown in . Here a diode with a nearly ideal barrier is compared with two others that have a certain amount of defects at the metalsemiconductor interface. As described in several papers (e.g. Defives et al.), at low forward voltage the areas with the lower barrier dominate the I-V curve. At higher forward voltage the areas with the higher barrier take over, because they are represented by a much larger part of the total interface. This results in a typical appearance of the log I vs. VF plot showing two or more separate linear branches each representing one specific barrier. The "perfect" diode, on the other hand, has only one linear region stretching over many orders of magnitude. One would expect from that diodes 2 and 3 will not exhibit a suitable blocking behavior, but this is not necessarily the case. As described by Tung et al., and discussed by Treu et al. in the specific case of SiC, such areas with lower barrier can be shielded under reverse bias by the space charge region of the surrounding high barrier regions, provided they are small enough (i.e. they have a diameter less than the depth of the space charge region). This is similar to the effect intentionally used in merged p-n Schottky diodes (Dahlquist et al.) to reduce leakage current. For a production process with reasonable yield, it is essential to keep all the unavoidable imperfections of the metalsemiconductor interface small enough that they cannot generate excess leakage current. In fact, we get a much better yield for our Schottky diode production than one would expect from the log I vs. VF plots for the diodes. High volume production Besides this basic understanding of the SiC Schottky diode, it was important for us to build a production line with a capacity in the order of several million devices per year. The line must be flexible enough to react to the highly unpredictable speed of market development, and must also minimize costs. The last point is especially important, because SiC devices already suffer from very high substrate cost as a barrier to market entry. Therefore it was our aim at Infineon to minimize the investment dedicated to SiC. We have been able to adjust our technological process so that we can mostly use standard 6 inch Si tools, with very few exceptions such as SiC epitaxy. We believe that there will soon be a demand for large volumes of the new SiC Schottky diode because it exhibits the following properties. A widely expanded voltage range SiC Schottky diodes offer a very low specific on-resistance with high rated voltages. Unlike Si and GaAs Schottky diodes, there is only a moderate increase in leakage current with increasing temperature (Treu et al.). The area-specific differential on-resistance of a 600 V SiC Schottky diode increases from about 0.9 m.cm2 at room temperature to 1.8 m.cm2 at 150 C. This positive temperature coefficient makes the Schottky diode well suited for paralleling without the risk of thermal runaway. When switching a Schottky diode off there is no need to remove excess carriers from the n-region as there is for p-n diodes. Hence no reverse recovery current will show up. Instead only a displacement current for charging the junction capacitance of the diode can be observed (see ). The current transient depends only on the external switching speed up to very high frequencies. The charge transported by this current is very low compared with the reverse recovery charge (Qrr) of pin diodes. Due to the different origin of this charge we have named it capacitive charge (Qc). Qc and the switching power losses of SiC Schottky diodes are not only ultralow. Compared to Si ultrafast diodes, where losses depend strongly on dI/dt, current level and temperature, they are more or less independent of these boundary conditions (). A dependence of Qc on these parameters cannot be seen at the same scale as with a benchmark Si diode approach. Again this is due to the capacitance-like behavior of this device in reverse direction. Active power factor correctors As always when trying to replace a cheap and highly current-overload-stable bipolar device with a unipolar device, the most striking argument is the need for high switching frequencies. There is one mass-market application in which this need is especially pronounced: the active power factor corrector (PFC), an example of which is shown in . The use of power factor correctors is growing strongly worldwide, driven by legal requirements. Boost converters are usually chosen to realize active power factor correction. They can be driven in discontinuous current mode (DCM) or continuous current mode (CCM). The DCM solution does not require an ultrafast diode like the SiC Schottky diode, but has several drawbacks: all circuit components have to be oversized because of the high peak currents, the system becomes unstable at light load and complex EMI filtering is necessary. The CCM solution lacks these disadvantages. The circuit components don t have to be oversized, the system operates stably at light load and the demands placed on the EMI filter are less rigid. However, the power losses in the MOSFET and in the diode due to reverse recovery dramatically limit the efficiency and switching frequency of a CCM boost converter (see ) when a conventional or ultrafast Si p-n diode is used. As SiC Schottky diodes do not show reverse recovery behavior, the stress on the MOSFET will be reduced due to a very low current spike during the turn-on transient. A less expensive MOSFET can be chosen and the entire system can be made more reliable. The efficiency is nearly independent of the switching frequency due to very low total switching power losses (figure 6), which is ideally suited to CCM. Boost converters can therefore operate at much higher switching frequencies. Reducing the boost inductors opens up new horizons in power density. This also has a strong impact on the cost of the inductor. EMI challenges at high frequency The EMI norm regulation begins at 150 kHz, so the main harmonic of the boost converter can be well inside this range. One would suspect complications achieving the EMI norm in case of higher switching frequency, but this concern is unfounded. Typically current-compensated double choke inductors have a maximum impedance (i.e. highest damping efficiency) in the frequency range from 300 kHz to more than 1 MHz. The main and higher harmonics of a boost converter running at 300500 kHz will be filtered with the maximum efficiency, so the increase in switching frequency to 300500 kHz does not cause a need for additional EMI filtering and electrical noise can even be reduced. These properties give circuit designers a new freedom when optimizing their PFC applications. Designers can increase the switching frequency, and reduce the size of passive components and semiconductor switches. They can also shrink or eliminate heat sinks and increase reliability and power density. Market predictions It is not enough to outline the benefits of the Schottky diode for this application; we also have to examine the size of the potential market. The real high-volume PFC market segment is in the power range up to 250 W covering large consumer applications like PCs and televisions. This several hundred million part per year market is extremely cost sensitive, making it an unlikely entry market for an expensive SiC Schottky diode. On the other hand, in the power-supply range from 200 to about 1000 W many high-end applications are positioned like servers, base stations for mobile phones and infrastructure for terrestrial communication. For this market reliability is very important. If one can reduce the risk of failure of the MOSFET switch in the PFC stage by replacing a conventional diode with an SiC diode, this will justify the still high price of the new diode. The same holds true in applications where high power density or ruggedness towards overtemperature are required. These high-end applications offer large-enough volumes to initiate and ramp up the production of SiC Schottky diodes. Access to the really huge volumes of the consumer market will follow with the associated price reduction. Further reading F Dahlquist et al. 2001 Materials Science Forum 338342 683 D Defives et al. 1999 IEEE Trans. Electr. Dev. 46 449 H Mitlehner et al. 1996 Proc. 9th EDPE (Dubrovnik, Croatia) 64 R Rupp et al. 1999 Mat. Res. Soc. Symp. Proc. 572 149 M Treu et al. 2001 Materials Science Forum 353356 679 R Tung 1992 Phys. Rev. B 45 13509 A Vrob ev et al. 2001 Materials Science Forum 353356 103