The Spirito Effect Improved My Design—And I Didn’t Even Know It

Oct. 20, 2011
Paul Schimel relates a personal experience with a class AB audio amp to switching supplies to explain the Spirito effect on MOSFET Safe Operating Area.

Fig 1. The lateral MOSFET is built horizontally on a p-type substrate (for an n-channel device). There is no clear, single PN junction from source to drain. The substrate is not the drain terminal of the device.

Fig 2. The trench MOSFET is built vertically on an n substrate that’s the drain terminal of the device (again assuming n channels). The gate is implanted into the silicon and can modulate more of the channel, yielding lower RDS(On) per unit die area.

Fig 3. The planar MOSFET is very similar in cross section to its younger brother, the trench MOSFET. The features tend to be larger and the gate can’t influence as much silicon as the implanted gate in the trench devices—a little less gain, a little less RDS(On)  per unit area, and a whole lot more backbone for those rough and tumble applications.

About 15 years ago, a friend stopped by with a really nice MOSFET-based audio amplifier that had a blown output stage. The devices in the amplifier were set up as a source follower with a nifty “trans-nodal voltage amplifier” driving them to basically rail-to-rail output capability and incredible slew rates. Thanks are due to David Haffler for the creation and the amplifier, but this one had seen some abuse and came to me as broken, three hours before it was needed at the gig.

I couldn’t discern, nor did I care to examine, the finer points of the output devices—didn’t have time. So I pulled the old devices, looked up the part numbers, and pondered what I had. I didn’t have the original devices, of course. Oddly, the body diode was left off of the schematic depiction of the device on the datasheet for the original output devices. The drain and source terminals on the pinout also were reversed. These two attributes seemed really odd. The magnitude of the voltage rating of the devices was 250 V for n and p channel alike. They were rated for something on the order of 16-A drain current.

I found some devices that required slightly lower voltage but should have worked. They were rated for 200 V and the supply rails in the amplifier were ±85 V or so. That was okay, really close, but okay. When I substituted them in, I verified the different pinouts and triple-checked the work. At power up and just after a quick bias point adjustment for 20 W or so of quiescent power on my power meter (on a 600-W channel), the amplifier died within a few seconds—dead short on all new output devices.

Dead On The Bench

So I did what anyone in a hurry would do. I went bigger and better. I pulled some 400-V devices from stock. They were rated for much more current. I quickly soldered them back in circuit, rebiased just barely into AB operation, and the amplifier ran great. It runs great to this day. It was a good gig. The band did well. My friend, on lead vocals, was happy with the sound. I went along and helped set up and run the boards.

So how does a pro audio gig and a blown amplifier relate to what we do in the switch-mode power-supply design world? Let’s get into that. In general, all MOSFETs are made of smaller cells combined in parallel to make one large device. The gates and sources of the cells are connected by various metallization layers on the top surface of the die. The drain is usually common to all cells and on the backside of the die.

The original devices in the clandestine amplifier were lateral MOSFETs. They were built and designed for linear amplifiers. Lateral devices have no intrinsic body diode, and the drain and source terminals are metalized on the processed surface of the device. This is a very different bird from the n and p channel MOSFETs that we use in the switching power-supply design arena where the substrate is the drain connection (Fig. 1).

The threshold voltages of the old lateral devices were much higher, the gain was much lower, the linearity was very well specified, and there was absolutely no information about on resistance (RDS(On)), Rg, or avalanche rating in the datasheet.

The devices that I had chosen to drop in were trench-type MOSFETs. Trench devices are set up for very low RDS(On), reasonable avalanche performance, and very uniform cell-to-cell capacitances. Trench devices are not set up for uniform cell-to-cell gains (Fig. 2).

If we look at what we are familiar with in a MOSFET, we know RDS(On) well. We know that when the device is driven into saturation, RDS(On) increases with temperature and threshold voltage drops. This means that the smaller cells within the MOSFET tend to heat up and share current with increasing temperature when they are saturated.  

In linear mode, this isn’t necessarily the case. When the channel is just barely accumulating charge and conducting, the behavior of a trench device is different. The trench device is designed so each cell sees exactly the same gate-to-source voltage and the same capacitances. What it’s not designed for is each cell having the same gain. If one cell has a higher gain than neighboring cells, it will conduct more current in linear mode, causing a hotspot.

Again, saturation overrides this mechanism entirely. In the cases that lie to the lower right-hand corner of the safe operating area (SOA) curve with higher voltage across the device and small currents, these hotspots will pass more and more of the current with increasing temperature. This is a runaway condition, akin to second breakdown in a bipolar junction transistor (BJT). This only happens in linear mode.  

When I dropped in the 200-V trench devices, I didn’t have the instrumentation to see it, or the patience to understand it, but the amplifier blew up based on very low currents at fairly high voltages—that of the bias point. This was hotspotting in the trench-type MOSFETs.

When I dropped the 400-V devices into the amplifier, they too were trench-type devices. I had just shifted the bias point to the left in the SOA curve of the device—back from the Spirito region. At 85 V, the 400-V MOSFET never saw hotspotting. It was safe. It worked well. The amplifier still works well today in spite of a decade and a half of road abuse. While it was a meatball substitution, it accomplished the goal. Total harmonic distortion (THD), intermodulation distortion (IMD), and square-wave performance were on par with that of the untouched channel.

The Spirito Effect

Some time later this effect was researched and documented by a researcher at International Rectifier named P. Spirito. Known as the Spirito Effect, it shows up in the SOA curve of the MOSFET as a steeper downward slope at higher voltages and lower currents. All planar and trench MOSFETs exhibit the Spirito effect to varying degrees. High-density trench devices are particularly impacted.

It is very important to understand that this Spirito Effect is a fairly new discovery in the world of power MOSFETs. Not all datasheets are created equal. IR datasheets reflect the Spirito region in their forward-biased SOA (FBSOA) curves. Other vendors might not depict this data as accurately.

A Reliable Ancestor

To get back to MOSFET structures for a moment, we need to discuss the predecessor to the modern trench devices—planar MOSFETs. There is good news! There is a family of MOSFETs that lies between the obscure lateral devices and the modern low-RDS(On), high-cell-density trench devices. These devices are planar MOSFETs.

Built on planar processing. Planar MOSFETs have larger cell pitches due to the limitations of the diffusion processing. They have lower gains, again due to the diffused profiles and the small area of the channel that the gate modulates. These two attributes make planar devices a fine choice for linear applications. They will have a more rugged FBSOA in linear applications (Fig. 3).

In general, when we consider MOSFETs in a power electronics design, we need to consider the application. For linear applications including linear amplifiers, servo amps, hot-swap circuits, and linear regulators, we need to look at the SOA curve of the device very carefully. Planar devices will lend themselves to these applications over trench-type devices.

For fast switching where avalanche or linear mode operation is not a concern, a trench device will offer the best attributes for the application. Whether you’re operating a dc-dc converter, commutating a motor, or perhaps running a linear regulator or a hot-swap circuit, IR has devices for the application. The downside is that the datasheets don’t directly tell which devices are planar and which are trench. For this information, you can contact your local FAE and chat about the application a little bit. We can furnish you with the best device for the job.


G. Breglio, F. Frisina, A. Magri, and P. Spirito, “Electro-Thermal Instability in Low Voltage Power MOS: Experimental Characterization,” IEEE Proceedings ISPSD 1999, Toronto, p233


To join the conversation, and become an exclusive member of Electronic Design, create an account today!