Njfet and Pjfet make good enough pairs to be extremely useful. I just got 3 awards in TAS for products made almost exclusively by K170/j74 jfets, and more are on the way.
Having lived with jfets for the last 45 years, I have seen their beginnings, and their gradual evolution, finishing with the Toshiba 2SK146/J73 pair that made ultra low noise, low open loop distortion designs possible. This part was mentioned in books and engineering articles on low noise design along with 'then revolutionary' new ultra low Rbb' complementary bipolars. All this by the early 1980's.
Now what is WRONG with these devices? There are at least 2 deviations that have to be taken into account: First is lower Gm for the P channel devices for the same size geometry. There are 2 ways of dealing with this, either resistive ballast to lower the Gm of the N device to 'match' the P device, or to parallel the P channel devices, either in the same package or multiple packages.
Most American parts designed complementary parts with similar size and therefore had to be balanced resistively, but the Japanese just made the parts with amazingly similar Gm's, but then the input capacitance went up 2-3 times in the P channel part.
For RF this might be a big problem, but for relatively low-Z audio work, the complementary design will win almost every time. It isn't necessarily because of the input stage, BUT the second stage which has fully complementary drive available. This makes for a more balanced voltage swing where it counts.
Because there is significantly more input capacitance from the P channel devices, it is best to cascode them as well to remove Miller effect based distortion. Either a straight or folded cascode will do, or else use a relatively low value load resistor, as we used to do with the Levinson JC-2 long ago, and the JC-3 sometimes discussed here.
The second major problem is with high output conductance in the P channel devices, so that they look more triode like on a semiconductor analyzer screen compared to the N channel 'complement'. Cascoding takes care of this as well. All in all, it works pretty good, even without lots of feedback to hide things.
Having lived with jfets for the last 45 years, I have seen their beginnings, and their gradual evolution, finishing with the Toshiba 2SK146/J73 pair that made ultra low noise, low open loop distortion designs possible. This part was mentioned in books and engineering articles on low noise design along with 'then revolutionary' new ultra low Rbb' complementary bipolars. All this by the early 1980's.
Now what is WRONG with these devices? There are at least 2 deviations that have to be taken into account: First is lower Gm for the P channel devices for the same size geometry. There are 2 ways of dealing with this, either resistive ballast to lower the Gm of the N device to 'match' the P device, or to parallel the P channel devices, either in the same package or multiple packages.
Most American parts designed complementary parts with similar size and therefore had to be balanced resistively, but the Japanese just made the parts with amazingly similar Gm's, but then the input capacitance went up 2-3 times in the P channel part.
For RF this might be a big problem, but for relatively low-Z audio work, the complementary design will win almost every time. It isn't necessarily because of the input stage, BUT the second stage which has fully complementary drive available. This makes for a more balanced voltage swing where it counts.
Because there is significantly more input capacitance from the P channel devices, it is best to cascode them as well to remove Miller effect based distortion. Either a straight or folded cascode will do, or else use a relatively low value load resistor, as we used to do with the Levinson JC-2 long ago, and the JC-3 sometimes discussed here.
The second major problem is with high output conductance in the P channel devices, so that they look more triode like on a semiconductor analyzer screen compared to the N channel 'complement'. Cascoding takes care of this as well. All in all, it works pretty good, even without lots of feedback to hide things.
Last edited:
Have you had much time to compare the linear system part of the future to Toshiba of the past. As to how the noise is and curves are ? Funny thing I have been listen to your design for 40 years my how time flies.
Yes it's just plain physics, the mobility of holes and electrons are different so N and P FET's can never be perfect compliments. If you fiddle the areas so the gm's match the capacitances are very different.
Because the Vbe equation is the same for both. In fact in the other details they are not usually very complementary. I've given up arguing that the diode equation is fundamental for bipolar design.
For example the process that most of the DSL drivers were done in had a 3X difference in Is, 2 or more times beta, VA and ft, and totally different behavior near saturation.
As one with no experience in the matching of npn and pnp devices I have a question. Most of what I think I see are matches that are made between the same manufacturer SCAxxx or whatever the numbers are. Are there ever times where the matches in electrical values are better by matching a transistor from let's say Toshiba and another from Fairchild where those match better than the same brand or is there something inherently better matching like devices from a single manufacturer?
- Status
- Not open for further replies.