Resurrecting the BJT: new SiC BJTs enable lower system cost in a PV inverter
05 February 2013
Recent work with silicon carbide (SiC) has given the BJT new life, producing a device that enables higher power density, lower system cost, and easier design-in.
When used in a PV power converter, a SiC BJT results in good efficiency and -- perhaps more important -- makes it possible to use smaller, less expensive components, for significant cost savings at the system level.
More than 30 years ago, CMOS alternatives like MOSFETs and IGBTs started replacing silicon-based BJTs in most power designs, but today, new versions, based on silicon carbide, are giving BJTs new relevance, especially in high-voltage applications.
The SiC format delivers higher switching frequencies at equivalent or lower losses, and produces higher output power in the same form factor. A design that uses a SiC BJT can also use a smaller inductor, and that can lead to substantial cost savings. The silicon carbide process produces a BJT that is more expensive than one based on just silicon, but the benefits of using SiC technology make it possible to save elsewhere in the design, for a lower overall cost. This article presents a step-up converter, designed for use in a PV conversion stage, that takes advantage of SiC BJTs to produce good efficiency at a significantly lower system cost.
The benefits of silicon carbide
There are several reasons why silicon-based BJTs fell out of favor in high-voltage applications. To begin with, low current gain in a Si BJT produces high driving losses, and these losses get worse as the current ratings go up. Bipolar operation also leads to higher switching losses, and generates high dynamic resistance within the device. Reliability is an issue, too. Operating the device in forward-biased mode can result in the formation of high-temperature filaments with high current concentration, and that can lead to device failure. Also, the voltage and current stresses that occur during inductive load switching can cause the electrical field stress to shift beyond the drift region, resulting in reverse-biased breakdown. This places strict limits on the reverse safe operation area (RSOA) and means silicon-based BJTs have no short-circuit capability.
New BJTs, implemented in silicon carbide, don't present the same problems. Silicon carbide supports a band gap that is three times wider than that of silicon, and that leads to higher current gain and lower driving losses, so the BJT is more efficient. The breakdown field with silicon carbide is ten times higher than with silicon, so the device is far less susceptible to thermal runaway and is much more reliable. Silicon carbide also does a better job at higher temperatures, so the application range is broader, and can even include automotive environments.
From a cost perspective, the high switching frequencies of silicon carbide make it possible to save at the hardware level. A BJT based on silicon carbide is more expensive than one based on plain silicon, but the high power density of the SiC process translates into higher chip utilisation, and supports the use of a smaller heat sink and smaller filter elements. Using a more expensive SiC BJT actually saves money in the long run, because the overall system is less expensive to produce. An example of this is the step-up converter we designed. It's intended for use in a photovoltaic system rated at 17 kW, with an output voltage of 600 V and an input ranging from 400 to 530 V.
The BJT's driver circuit presented an opportunity to reduce loss and improve system efficiency. The driver does two things: it charges and discharges the device capacitances quickly, to enable fast switching, and it ensures a continuous supply of base current to keep the transistor saturated in its on-state.
To support dynamic operation, a driver supply voltage of 15 V results in faster transients and improves performance. The threshold voltage of the SiC BJT is around 3 V. There's typically no need for a negative driving voltage or Miller clamp to increase immunity.
The SiC BJT is a "normally-off" device that is only activated when there is a continuous supply of base current. Selecting the value of the base current for static operation involved balancing a tradeoff between conduction losses and driving losses. Driving losses are still important with an SiC BJT, despite the higher gain values (and thus the lower base currents), because the wider band gap of the SiC format necessitates a higher forward voltage between the base and the emitter. Doubling the base current from 0.5 to 1 A lowered the forward equivalent resistance by only 10%, so we needed to reduce conduction losses while also shifting the saturation to high levels. This was an important consideration for our step-up converter, because it operates with higher levels of current ripple. A base current of 1 A increased the switching capability to 40 A.
Static driving losses are a function of the selected driving voltage and the input voltage (which indirectly implies the duty cycle value). A driving voltage of 15 V, needed to enable high switching speeds, produces a loss of nearly 8 W, mainly concentrated on the base resistor. To offset this loss, we used two separate voltage supplies for dynamic and static operation. The schematic is given in Figure 1. The control signal to the high voltage driver is "chopped" so that it's only enabled during the switching transient. The static driving stage is fed by a lower voltage, which reduces static losses, and remains activated during the entire turn-on time.
Reducing the filter size
Operating at a higher switching frequency reduces the cost of the passive elements. To increase the power density further, the group looked at ways to improve the filter inductor. After reviewing the capabilities of various core materials, a newly available cut-out core made from Vitroperm 500 F, a thin-laminated nanocrystalline material as chosen. The material produces low loss levels and good operation at high frequencies. It also operates with high values of saturation flux, which means it can be much smaller than a comparable ferrite core (right side of figure 2). Using the Virtoperm core resulted in a filter inductor that is about one quarter the size of our reference system.
Figure 2 depicts a factor representing the inductor size as a function of the switching frequency, for different materials under a maximum level of current ripple (40%). Here it was assumed that inductor volume can be approximated by the value of the inductance, which in turn depend on the peak flux density and switching frequency. After reaching a specified critical point, with specific losses defined at 100 mW/cm3, the peak flux needs to be reduced to avoid overheating so that operation beyond such point results in no significant size decrease. Vitroperm 500F enabled for the selected frequency the best performance among all materials.
Figure 3 gives the measured efficiency levels, including driving losses with the two-stage solution. A distribution of the losses, according to the calculations, is presented below the curves. The system can reach high current loads without reaching critical temperatures or saturation. The two-stage driving solution lowered driving losses to about 0.02% of input power. The lower level of overall losses reduced the size of the required heatsink and the higher switching frequency enabled smaller filter elements. All these features contribute in the end to lower system cost.
Silicon carbide has given BJTs a new lease on life. Unlike their Si-based forebears, SiC BJTs deliver low conduction losses, a high breakdown field, and stable operation over a much wider temperature range. Using two supply voltages in the driver circuit reduces driving losses and produces good efficiency. The higher switching frequencies make it possible to use a much smaller inductor, and that leads to significant cost savings at the system level. In high-voltage applications such as PV inverters, which benefit from high power density, lower system cost, and easy design-in, SiC BJTs present an attractive alternative.
Figure 1. Using two supply voltages reduces loss
Figure 2. Inductor size for different core materials as a function of frequency, and comparative sizes of Vitroperm and ferrite cores
Figure 3. Efficiency at 48 kHz with driving losses, and a photo of the prototype
Contact Details and Archive...