Superjunction sparks super devices in silicon carbide
Armed with a clever charge-balance structure, SiC power devices are pushing beyond their limit.
BY JAN CHOCHOL AND ROMAN MALOUSEK FROM ONSEMI
The entire power electronics industry is buzzing with SiC activity. It’s no wonder, given that SiC devices have a dielectric breakdown field strength that’s ten times higher than that of silicon, as well as twice the electron saturation velocity of the incumbent, three times its bandgap, and a thermal conductivity that is better by a factor of three. All high-power applications benefit from these traits, shown in Figure 1, so it’s of no surprise that vendors, fabs and OEMs are embracing the rapid adoption of SiC. All are trying to outperform, outlast and outclass their competition.
Figure 1. Areas of application for different power electronic technologies.
Due to all this frenetic activity, there is some turmoil in this sector, creating opportunity for all kinds of development. This has motivated our team at onsemi to explore various concepts. Against the backdrop of a maturing SiC industry, we have been able to revisit some topics that just a few years ago would be either unfeasible, too expensive or just plainly not possible with the toolset available at that time.
One such concept is the superjunction. Regardless of design, any device that is made from SiC will benefit from its unique properties. They include a large electric breakdown field that shaves tens of micrometres of epi thickness, greatly reducing the on-state resistance for a given voltage rating compared with silicon. But with the superjunction structures that we have explored, one can go a step further. Generally, this structure consists of alternating regions that are highly doped, tightly spaced and with equal and opposite doping – an architecture that ensures charge balancing. With such a structure, the high doping results in superior conduction that reduces resistance. But that’s not the only benefit we get – operated under reverse bias, the superjunction is fully depleted, with the electric field spread evenly in a roughly rectangular shape. Thanks to this, compared with a classical unipolar drift region, where the electric field must be trapezoidal (see Figure 2), we realise a higher breakdown field at the same drift thickness.
Figure 2. Schematic diagram of JBS diode with unipolar drift region (left) and superjunction JBS diode (right).
The real question is how can we make such a structure? We need to have defined pillars, with a controllable concentration reaching several micrometres below the surface. As each micron of superjunction depth can block about 200 V (we go on to show this), a 1200 V device needs more than 6 μm of well-controlled superjunction pillars.
One option for forming the pillars is ion diffusion. However, those working in the SiC community don’t tend to concern themselves greatly with this technique, as there is simply no way to diffuse ions in SiC. This has led some groups and companies to try other approaches, such as trench filling by CVD.
Revisiting ion implantation
But in our pilot study, we decided to take another look at ion implantation, due to the simplicity of the method and the availability of tools. We have investigated two different techniques. One involves using very high energy implants. Roughly speaking, this is a brute force approach; we push the implants deep into SiC, due to their high kinetic energy. Higher energy means deeper penetration. Then, using multiple implants, we can chain the concentration profile together to create a seemingly uniform distribution. The caveat of this approach is that we have to shield the other pillar from the implantation. So one implant can be blanket, while the other must be masked and have twice the dose of the other. Success is not easy, as there is only so thick a photoresist that can block the implant.
Figure 3. JMOL simulation of 4H-SiC lattice. Blue arrow shows the direction of c-axis of the crystal. (a) view through the channelling direction; (b) sideview; (c) tilted lattice; (d) tilted lattice sideview.
Due to this limitation, we have also explored the channelled implant. One key characteristic of the 4H polytype of SiC, used for power applications, is that it permits ion implantation along a preferential angle, where ions ‘see’ through the lattice and the number of ion-lattice collisions plummets. This principle, illustrated in Figure 3, leads to a concentration profile that is far deeper and flatter than that normally possible with a random implantation direction. For example, a 900 keV non-channelled implant of aluminium reaches a depth of up to 0.7 μm, while channelling with this energy extends the depth to 3 μm.
With this approach we have produced pillars by combining a blanket high-energy nitrogen implantation, which does not require a mask, with a masked channelled aluminium implantation. The success of this scheme is seen in profiles obtained by secondary-ion mass spectrometry (see Figure 4).