Dear Bob: I'm running into a head-scratcher. I am working on a 2-µA current source (pnp transistor plus sense resistor in the emitter lead, and an op amp to control the pnp's base so the voltage across the sense resistor is constant) with as much compliance to use as much of the 5 V as I have available.
(Constancy of current is important but I don't care too much whether it's exactly 2 µA as long as it doesn't have much tempco/drift.)
But I've only found one pnp transistor (PN5138) with any beta specs for collector currents that go down that low. (The PN5138 has—I saw—a guaranteed spec of 50 minimum at 100 µA. I did not see any guaranteed spec at 1 µA. This transistor is from the Process 66, which is not nearly as high beta as process 62. Good beta at low currents like 1 µA is a characteristic—not a guaranteed spec. /rap)
Any idea why? I can see the manufacturers wanting to test/characterize beta at the high end, but why are they leaving off the low end? (It was very uncommon, even 20 or 30 years ago, to see guaranteed specs for beta at 10, 1, 0.1, 0.01, and 0.001 mA. Almost nobody wanted to pay for that. The 2N930s and 2N2484s were npn types, and they were fully specified—at high prices. /rap) Nobody seems to support discrete small-signal bipolar devices anymore. (They never did want to. If you wanted to buy 2N4250s, 2N2605s, or similar pnp devices in mil-spec format, for $20 to $40, you could do it. But for your application, just buy some pnp transistors with beta = 200 min at 1 mA. All modern high-beta pnp types will have good beta hold-up at low currents. Get something like the old 2N4250, PN4250, 2N2605, or something similar. If you buy them at Fairchild, they would be anything with high beta such as 300 min at 1 mA, or 250 at 100 µA, and they would be from Process 62. That was what they were called when NSC made them. /rap)
I'd use a PFET instead of a pnp, but I'm told FETs begin to leak at high temps with VGS = 0 (e.g. for NFET, 2N7000 specs ≤1 µA at 25°C, VGS = 0, and ≤1 mA at 125°C, VGS = 0). And I don't have enough voltage available to apply a reverse bias to VGS. Any thoughts?
Jason Sachs (via e-mail)
Pease: FETs are not guaranteed to not leak, but they certainly are not guaranteed to leak. The manufacturers just don't want to test below 1 µA of leakage, as that takes time, which they hate to waste. Those Process 62s will make you happy.
Dear Bob: I was looking at some back issues and I came across your column on the selection of tubes in a Philbrick K2-X (electronic design, Oct. 18, 2004, p. 16). It made me recall an experience of mine. Way back in the middle of a long-gone century, I was designing a VHF television tuner. I was using a dual tube, I believe a triode-pentode, specified by the manufacturer for that purpose. We used a major brand, adjusted the circuit, and found that it would work properly with any tube they sent us.
Then we tried another, equally reputable brand. No go. They did not work properly. Both were specified for tuners. They had the same published specs. But the specs defined only the operation of each section. They did not specify the physical layout. It turned out that the orientation of one section was the same for both companies. But the orientation of the other section was different. This caused a difference in the mutual inductance between the elements of one and elements of the other. At 200 MHz, mutual inductance was crucial. (Uh, yeah! /rap)
This indicates the danger of relying on specifications that cover only the major requirements of a product. These are adequate for most applications but omit "minor" details that may be crucial for your particular application. On occasion I have encountered the same problem with mil-spec integrated circuits and generic prescription drugs.
Matthew W. Slate (via e-mail)
Pease: You're exactly right, Matthew. There are many factors that are not trivial.
Comments invited! [email protected] —or:
Mail Stop D2597A, National Semiconductor
P.O. Box 58090, Santa Clara, CA 95052-8090