IXTK90N25L2, or how I bought used transistors on Taobao

While working on the power stage for my new electronic load, I got the idea of using proper transistors to dissipate the “excess” energy, and IXYS’s IXTK90N25L2 caught my eye. But new ones from a reputable source cost a serious amount of money, so when putting together another Taobao order, I decided to buy some.
More details after the break.


First, I should probably explain why these transistors in particular, rather than the usual IRFP250–260.
Besides their simply excellent specs, the datasheet for these transistors mentions a seemingly obscure thing called FBSOA, and especially Guaranteed FBSOA at 75.


You can read about SOA (Safe Operating Area) on Wikipedia, though it’s in English. I’ll try to explain it in plain language.

Many power MOSFET datasheets include a chart labeled Safe Operating Area. It shows the region in which the manufacturer guarantees stable operation of the component. But not all MOSFETs have a DC region, and even when they do, it’s usually very narrow.
There’s not much point looking at the SOA chart at 25°C: to create those operating conditions, you’d need a cryogenic setup to keep a real transistor at 25°C while it dissipates a lot of power. So let’s look at the chart on the right. It’s a little more conservative, but it’s specified at 75°C, which is more realistic. I keep the temperature in my loads below 90°C.
In the lower region marked DC on the chart, you can see that the transistor is theoretically rated to operate in linear mode (DC) at 100 V and 10 A (1000 W), or 200 V and 3 A (600 W), or 30 V and 20 A (600 W).


Now let’s look at the SOA region for the IRFP250. What do we see? There’s no DC region at all—only for switching operation, and that’s it!
Of course, you can use it in linear mode in electronic loads, and people often do, but in that case they simply limit the power to 50 W for the TO247 package or 30 W for TO220.


So why does this happen?
Below, I’ve tried to explain the internal structure of a bipolar transistor and a power MOSFET “with my fingers,” as we say in Russian.
Inside a power bipolar transistor, everything is just like in a low-power one, only “bigger.” That’s why bipolar transistors usually have the same power dissipation rating in switching and linear modes. Strictly speaking, a bipolar transistor’s switching mode is its saturation mode, and it performs worse there than in linear mode, because a bipolar transistor switches faster in active mode (without saturation). But that’s a whole different topic; we’re talking about the differences here.
Inside a power MOSFET, things are completely different: it’s essentially a bunch of small transistors connected in parallel. To get more current and lower resistance, you connect more of these transistors. Fortunately, connecting MOSFETs in parallel is very easy, unlike with bipolar transistors.
This all works, and works great, but... in switching mode, where the transistor is either on or off. This approach barely works in linear mode, as in electronic loads and linear voltage/current regulators. The reason is that although the transistors on the die are very similar—they’re made on the same wafer in the same process—they’re not identical. As a result, their characteristics vary, and some transistors may turn on slightly earlier than others, taking most of the thermal load.

This doesn’t mean that if you use the same IRFP250 at 100 W, it’ll immediately die. Not at all. First, the smallest but most “active” transistor will fail, then another. Then you’ll reduce the power, and the load will work again for a while. Later, you’ll raise the power above a relatively safe level again, and a few more transistors will fail—there are a great many of them in there, after all. But sooner or later, the process will turn into a runaway failure, and you’ll end up with a shorted transistor. So, to reduce the chances of this happening, don’t exceed the values listed above.
In switching mode, things are simpler: the load is distributed more evenly among the elements, and everything works great.


Next, I’ll talk about the transistors themselves, as well as the ins and outs of buying them and the problems I ended up with, even though I tried to do everything right.

Why Taobao? It’s simple: you can sometimes find desoldered transistors there, and they cost noticeably less than new genuine ones. The listing shows a price of $1.50, which is pocket change compared with $25 at Mouser/DigiKey or $30 from an offline supplier


If I only needed one or two, I’d simply buy them locally. But I’m planning to build a high-power load, and needed 12–16 of them (4×3 or 4×4), so I decided to buy them on Taobao.
Knowing that when you buy new components in China that are supposedly genuine, the chance of getting counterfeits is about 99.(9)%, I immediately looked for used ones, i.e. salvaged from circuit boards.
I had no issues with the forwarding agent at all, but the seller packed them in such a worn-out bag that I was afraid to even pick it up.


Besides the transistors, there was some other junk in the bag, so they really did look used. I ordered 17: 12–16 for testing under load, maybe 12, and 4 “spares” or “in case some are defective,” plus one to break open and inspect the insides.


They were heavily used, with bent leads; some were broken off almost flush with the case, and a few still had bits of board plating on them. I wasn't buying them for display, though, so I didn't care. For a 10–20× price difference, I can live with that.


The transistors were also very dirty. Above, I tried cleaning a few with alcohol; the cotton swab turned black instantly. For some reason, though, the marking on one transistor became even harder to read. No matter.
Most of the transistor leads were simply broken and short, but one had a wire soldered onto it, and another had almost intact leads. I only got one like that :)
The underside has a large base plate, just as shown in the datasheet, but more on that later.


A couple of the transistors even arrived with damaged cases. Apparently, they were “removed” from the boards with a crowbar.


The case matches the original; you can even see the holes on the plate side above, which are also shown in the datasheet.


The transistors certainly look impressive. Here they are compared with more common ones.


They also weigh about the same as the originals, which are listed at 10 grams, though these have slightly shorter leads.


Everything seemed fine: the transistors were clearly used, and their appearance suggested they were genuine. But after the very first test measuring their on-resistance, they split into two distinct groups, plus two more.
1 pc — 0.1 Ω.
9 pcs — 0.18 Ω
2 pcs — 0.4–0.5 Ω
5 pcs — 1.9 Ω

At this point, I had no polite words left, only the unprintable kind...


The thing is, a good transistor should have an on-resistance of about 0.036 Ω, or 36 mΩ.


Interestingly, within each of the two large groups, the resistance was stable, with only minor fluctuations: around 0.178–0.182 Ω and 1.85–1.95 Ω. In other words, the groups were very clearly defined. Below are photos of the four groups in the order listed above.


The gate capacitance should be about 2140 pF. In reality, only the transistors with 0.18 Ω and 0.4 Ω resistance had values anywhere near that.


I ran the next test mostly out of curiosity, since the transistors turned out to be quite sensitive and would partially turn on even from the multimeter's continuity-test voltage.
Resistance (shown at the top left on the milliohmmeter) versus gate voltage (shown on the multimeter display).

First, the ones with 0.18 Ω resistance.


Now the ones with 1.9 Ω resistance.


Here they are in a table: the 0.18 Ω transistors on the left, and the 1.8 Ω ones on the right.


The graph came out a bit awkward: it starts at 3 V, and the grid at the bottom is marked in 0.5 V increments.


As you can imagine, I got curious about what on earth was going on, so I decided to open up one of the transistors. To do it “quietly and without making a mess,” I used a small vise. Maybe you had one of these too—they're incredibly handy :)

First up was a transistor with 1.8 Ω resistance. I was going to throw it away anyway.


Inside is a plate with the die. Unfortunately, most of it stayed embedded in the plastic. I really wanted to look at it under a microscope, like I did with flash memory chips.


The spot on the base plate where it was glued.


And in the plastic


The base plate really is copper: not only is it non-magnetic, it also looks like copper. Thickness: 2 mm.


When I tried to carefully split the case or somehow pry the plate out, it simply broke in half.


I don't think there's any point in looking at it under a microscope; these are just macro photos.


I decided to handle the second transistor differently. Knowing the die dimensions, I first marked out its area, then cut out a little square with a rotary tool, pried it up with a screwdriver, and ended up with the same thing as before.
The die measures 6x6 mm in both cases.


All right, I thought, I'll open up the third transistor, the 0.1 Ω one. I had a spare anyway.
Hmm... this one is completely different.


For starters, the die is a different size. The first two have a resistance of 1.8 Ω; the third is 0.1 Ω.


But there's another difference too. Didn't notice? The plate has a completely different stamping pattern, both around the die and in the flange area.


Okay, let's take the fourth transistor, the 0.4 Ω one (I had two of those), and open it up.
Well, this is a third design...


After opening them up, I found three different internal designs, differing both in die size (5x5 and 6x6 mm) and in the stamping pattern.
From left to right: 0.4 Ω, 0.1 Ω, 1.8 Ω


So, friends, what have we learned? These transistors are certainly impressive, but what we actually have is... a complete mess.
To avoid buying counterfeits (unless you pay a premium—and even that isn't a guarantee), I did everything I could. I found a lot of used transistors. They definitely look used, because going to this much trouble to make counterfeits... well, I don't know.
There are a few possibilities
1. The transistors are counterfeit. Somehow, I find that hard to believe; they really do look used.
2. The transistors are genuine, but then what's going on with their specs?
3. The transistors are genuine but damaged. They were used in some high-power unit that gradually failed, then was scrapped and dismantled.
4. The transistors are counterfeit, but were removed from some device. Here's the scenario: one crafty Chinese seller sold a batch of fake transistors to someone, who soldered them into a unit, realized they were fake, scrapped the unit, and another Chinese seller bought it and sold it off in parts.

What do you think, folks?
Based on my unfortunate experience, it turns out that being used—and even having traces of metallization left on the leads—doesn't prove that components are genuine at all. This experiment cost me about $26 for the transistors, $1.88 to ship them to the forwarding agent, and $4 for the actual shipping, or about $32 in total.
Only later did I notice that the order listed 16 transistors, even though I wanted 17 and received 17. How did that happen? A mystery...


That's all. I hope you found this review useful.