"Real World" JFETs

[JDoyle]

[I had started to reply to the post about the 2N5457/2N5458 JFET differences, but realized that this was probably info best brought out to its own topic.]

Even if you just started building effects this morning, and have been able to spend just an hour on this site doing some research, there is a good chance that you have already learned the Cardinal Rule of JFETs:

JFET parameters are highly variable from one individual part to the next - even for individual parts with the same part number, the same manufacturer, and even from the same batch.

This one fact influences EVERYTHING related to JFETs, from the way they are treated in educational and research literature, to the way they are assigned part numbers, to the actual price they command in the marketplace.

Because of the high level of variability, it is not only next to impossible, but damn near completely useless, for an author or publishing scientist to focus on a specific type or part number as a 'representative' example of a JFET. Simply because even if the reader were to acquire the part number used by the author, there is a much larger chance that those parts would NOT have the same characteristics as the one used in the article than they would. A best case, blue skies, 'today's the day to buy a lottery ticket' estimate would be that 20% of the JFETs, 1 in 5, would be in the same characteristic ballpark - and again, that is if they can be found at all, which is not a small obstacle, by any means.

Because of this, JFETs are almost always treated in the theoretical realm when their properties are discussed - everything is a variable (Idss, Vgs(off), Vp, gm, etc.) and then those variables are related to each other through equations. Unlike the case of BJTs, where there are most definitely still variable parameters (most obviously beta, but also the Early Voltage, among others), but those parameters can either be turned into actual numbers with a fairly high relation to practical realities (Ve=Vb-0.6V, Vt=0.026V, etc.) or ignored altogether without severe consequence to the design, there is only ONE solid number that can be found for all JFETs - that bringing the gate more than 0.5V above the source forward biases the gate-source diode - which is a situation we are always implored to avoid. So the one actual number that we have is not only old news (it's just a humble diode) but is a limit of operation, not a foundational function of operation, so we are left with the variables and completely bereft of any solid, real-world values until we have tested the individual parts. A secondary effect of this is that when authors delve into the circuits possible with JFETs, not only are the JFETs represented with variables, but so are the resistors used in biasing and as the loads - other than the 'standard' 1 Megaohm gate resistor. This is again a direct relation to the variability of JFETs themselves - you can't specify the value of a resistor if the current through it can be any value in a 5:1 window - each individual circuit becomes almost unique. A tangential moral of all of this is if you see a schematic of a circuit that contains single JFET stages with specific values for the resistors needed, be very, very, very, wary - and in my opinion, expect it to not work at all as drawn - because there is an 80% chance that it won't. 

The thing is - none of this is really the JFET's fault; not that they care. The fault lies almost entirely with the way JFETs are manufactured - but you can't blame a JFET for being a JFET. Because of the 'simpleness' of their structure (they are basically 'siamese twin' diodes - they share a body (anode), the gate, but have two heads (cathodes) - the drain and the source), there is very little that can be done to change the situation - especially when you consider that BJTs and MOSFETs are not only less variable parameter-wise, to the point of being nearly identical in the case of BJTs of the same part number, but are also much cheaper to produce (the gate of a MOSFET is 'grown' with a blast of steam or air). In reality, JFETs find themselves exceeding the abilities of both of the former types rarely, and then only in extremely exotic and niche situations that don't require the quantities needed to push the semiconductor industry to improve current designs or come up with new ones.

I know the manufacture of semiconductors bores a lot of people to death, but at the same time I think a quick discussion helps in understanding. Plus, there are a lot of similarities between it and the way that we make our own PCBs so it shouldn't be that far of a stretch to understand. Don't worry I'll keep it brief. Instead of starting with a blank copper PCB, the process starts with a chunk of silicon, either n or p type, but the opposite of the type that the JFET will be named - p type for n-channel JFETs, n type for p-channel JFETS. (JFETs were originally made from germanium, just like BJTs, but once it was possible to acquire relatively 'pure' silicon, as we all know, germanium went the way of magnetic tape - used only in situations where it's specific abilities make it the only choice; though recently there has been a fair amount of literature on the performance of germanium JFETs at temperatures approaching absolute zero, which, unfortunately, means outer-space, and therefore does not equal 'cheap' or even 'available through retail' if they do make a comeback.) Then, just as is done to the blank copper PCB, a pattern is made on the surface of the silicon via a photolithographic process (the 'advanced' version of PNP Blue). This pattern determines the internal structure/shape of the JFET channel. Then the silicon is placed in an oven and a gaseous form of a p or n type element (the type is opposite of that of the original chunk of silicon) is injected into the oven for a specified amount of time. At the high temperatures used, the silicon ceases to be a semiconductor and it becomes 'adjustable'; the atoms from the gas 'diffuse' into the areas of the chunk of silicon that are exposed (not covered with the pattern), adding extra electrons or holes and thus turning those areas into the opposite type of silicon: n type to p type (when holes are added) or vice versa. So while we immerse the copper board in a solution that removes the exposed copper and leaves copper under the pattern, a JFET is made by ADDING atoms to change its original polarity/type - but both use a pattern to control what is changed. This first diffusion creates the drain-source channel. Next an area of the channel is itself marked off by the same photolithographic process and a portion of the channel reaching from end to end is then diffused BACK to the original polarity type of silicon as the raw chunk that originally went into the oven - this last diffusion, along with the back, undiffused portion of the original chunk, are both gates. The channel is then separated from the gates by the pn junctions created by the diffusion and the depletion regions created by adjacent pn junctions. The two gates are connected together with wire to form a single 'gate' and contacts are attached to it and to the opposite ends of the channel, making the 'real world' gate, source, and drain contacts that we solder in place in our circuits.

The problem of variability arises because of the diffusion process. First, there is no such thing as a 'pure' element - there will always be impurities - which means that the result, the amount of p or n type changed to the opposite, will be a function of the amount of impurity in the sample (or, its 'impurity profile') - which is non-uniform throughout the sample leading to different parts of the silicon having different outcomes in terms of the amount (actually: depth) of one type changed to the other. Second, there is a limit to the accuracy of the photolithographic method. Anyone who has ever tried to iron on a PCB pattern that has too small traces or pads knows exactly what this means. Third, while the process of diffusion is a highly technical, scientific, and extensively researched subject, it is simply not possible to precisely control when the diffusion process stops. Anyone who has ever wondered if the etching of a PCB is ever going to finish (if the acid is old and loaded with copper from old PCBs, this is akin to a chunk of silicon with a high amount of impurities) or has left it in too long, eating away traces, experiences something similar. Gases already diffused continue to diffuse as the sample cools, even after the gas has been removed. Where the parallel ends with our homemade PCBs is while we can check the progress with some tongs, or rubber gloves, few people can do the same to molten silicon in an oven at well over 1000 degrees. All of this means that the depletion zones which control the current through the JFET, and are the essence of 'JFET action' (the field effect - the amount the depletion zone extends into the channel), can not be accurately controlled or reproduced. This is entirely different from BJTs where the controlling aspect, the forward biased base-emitter pn junction, is very well defined (because of physics) to the point that a 'guess' of 0.65V, and using the 'standard' leakage current value will allow for very close calculations. In a JFET, the depletion zones are a function of the diffusion process and the impurity profile - neither of which are readily determined, and certainly not as accurately as the forward voltage of a pn junction.

The end result of all of the whole process? The semiconductor manufacturer doesn't know what properties the JFETs have until they are cut apart and individually tested.

Now, something which I have written here once before, but I think is a little known aspect of JFETs, is the fact that a manufacturer doesn't set out to make a batch of J201s, 2N5458s, or 2SK117s. They use a 'process' - which is the combination of the photolithographic pattern and the schedule/timing/type of diffusion process. An individual JFET doesn't get a part number until the very end - after it has been tested and it's parameters are determined. In fact, for the American and (I think) European numbering system one 'process' can produce several different part numbers and those part numbers define the characteristics of the individual parts. For example, part numbers J201 and J202, as well as at least a half dozen other parts, are made in the same process, and quite possibly at the same time - they are chips that come out of the oven sorted by the Idss parameter (Vgs(off) tracks with Idss, so it means the same thing in terms of sorting, but it's much easier to measure Idss - connect the gate to the source and apply a sufficient voltage across the drain and source - than to measure Vgs(off), which is much more difficult to measure in a practical sense as it tapers off so gradually and low currents become hard to measure or differentiate from leakage currents). So a J201 is a J202 with a lower Idss/Vgs(off), a 2N5458 is a 2N5459 in the same manner - though there is overlap between the two making it possible for the opposite to be true as well. (Isn't this fun?) But hey - at least we get that much, and there isn't just a single 'J201' with an even larger range of possible values.

The Japanese numbering system for JFETs is different. The chips are tested and then assigned an Idss suffix/classification (O, Y, GR, BL, or V; lowest possible Idss to highest - not all are available for each part) and then assigned a part number according to their packaging - for example, a 2SK370 is the same pattern/process as a 2SK170 but encapsulated in a 'mini' package whereas the 2SK170 is encapsulated in a TO-92. Also, they can be, and often are, sorted for specific attributes such as noise (which in a JFET is mostly a function of the 'impurity profile' and therefore pretty much beyond control other than by a final testing/sorting) - a 2SK364 is a higher noise version of the 2SK170 - or rather they probably all started out as 2SK364s and then the lower noise chips were separated out and named 2SK170s. However, if you take a quick glance at the datasheets you may not believe me - the 2SK170 and 2SK364 appear to be different at first glance. But a close inspection of the limits of the characteristics and the graphs that follow reveals that the two are functionally equivalent and differ in only the addition of a 'min' value for gm in the 2SK364 and a different presentation of the information in a few graphs. (As a side note, a really good indicator of whether one JFET is the same process as another is the listed parasitic capacitances, Ciss and Crss, which are primarily a function of the pattern used in the gate diffusion - which is the more complex and characteristic-defining pattern - if they are listed as the same, then they are almost certainly made with the same process.) The reason for the cosmetic difference in datasheets? My guess is that the manufacturers don't want a prospective buyer to find out that if they can handle a little more noise they can purchase a cheaper part and get the same performance.

The upshot of all of this needed testing and sorting is added cost to the end buyer - it takes money, time, and equipment to test each individual part and that gets passed on to the customer. Add into that some desire of end customers that ends up being an uncommon occurrence or outlier, for example, a low noise, high Idss, part such as the 2SK170V (used in a lot of hifi stereo designs) and the laws of supply and demand rule, as always, and make that part expensive.

Hopefully this cleared up more questions than it raised!

Regards,

Jay Doyle
   

[R.G.]

Good one, Jay. Maybe with repetition, it will sink in.

One extension. Where yous say:

Because of the high level of variability, it is not only next to impossible, but damn near completely useless, for an author or publishing scientist to focus on a specific type or part number as a 'representative' example of a JFET.

I don't think you go far enough. I would say that it is more than near completely useless - it is irresponsible and misleading unless the author forces the reader to look at the caveats. 

[frank_p]

Thanks Jay, for all these explications !

[Paul Perry (Frostwave)]

And that is why I never use a fet for anything but a buffer or a switch, no matter how many intriguing circuits there are.

[drk]

reading this topic kinda gives the idea that we shouldnt use jfets... but if they exist, they gotta be good for something no? 

what are the advantages(versus bjt for example) of jfets then?

[Gus]

The basic thing is if you want to be repeatable with jfet builds understanding jfets is needed.

Different companies use different process numbers  IIRC a NS P50 is like an Interfet 26.  In the 1982 NS transistor book I think I counted over 25 fet numbers based on the P50,  case, pinout and selection of parameters. 

[brett]

Hi

what are the advantages(versus bjt for example) of jfets then?

Some of the good things are good: exceptionally high input impedance (hence their use in TL072 and many other op-amps).  I think the input impedance of a TL072 is something like a thousand billion ohms, so the inpedance of any circuit using such a device is set by other devices (such as the biasing resistors).

Also, some of the "bad" things are good for us stmpboxers:  the non-linearity can be a blessing.  When overdriven, many JFET circuits produce significant amounts of pleaasantly harmonic distortion.  The mu-amp is an example.  For details of making JFETs "work", I suggest reading RG Keen's "Fooling with FETs" article.

Have a good day.
Brett

[R.G.]

reading this topic kinda gives the idea that we shouldnt use jfets... but if they exist, they gotta be good for something no? 

Everything's useful for something - even if only a common object of scorn or a bad example.  

what are the advantages(versus bjt for example) of jfets then?

JFETs are simpler to make than BJTs; they have a higher input impedance, smaller capacitances and hence higher frequency response in general; self-limiting currents; lower noise in many situations. They are actually widely used, just not as discrete devices. The entire National Semiconductor line of LFxxx opamps and the TL06x/TL07x/TL08x and many other opamps use JFETs as the input devices for the opamp. In such applications, the JFET's variation is of minor consequence, since it's always hidden by the overal feedback.

Industry has largely abandoned discrete JFETs except in very limited circumstances where they can be selected, matched, or adjusted into operation.

A word about adjusting or trimming as the industry views it. Trimming is a Bad Thing. It's what you do when you can't do anything else because it takes time, and an adjustable part that can be misadjusted or drift. Most electronics is simply too cheap to afford trimming; it's not a viable alternative.

[Zben3129]

reading this topic kinda gives the idea that we shouldnt use jfets... but if they exist, they gotta be good for something no? 

what are the advantages(versus bjt for example) of jfets then?

You can get some great sounds out of them! Take for example the BSIAB2

I like to think of them like germanium transistors. You can't buy just any 2 germaniums and expect them to work correctly together. You must buy more than you need, interview, sort, mix and match, adjust bias, etc. Same deal with JFETs. The reason is also the same, extreme variation within a part number.

But where would we be without the great sound of germaniums (and JFETs...Phase 90, BSIAB2, Orange Squeezer...)


Zach

[DougH]

A couple things:

And that is why I never use a fet for anything but a buffer or a switch, no matter how many intriguing circuits there are.

Yes, I've gotten away from using them a lot as well. But I can't help but return to them on the breadboard from time to time because they are fun.

You can get some great sounds out of them! Take for example the BSIAB2

Be careful of mixing apples and oranges. A mu-amp is a self biasing circuit that by definition will mask some of the variability issues that JFETs have. As typically used in pedal circuits, JFETs exhibit the biggest weakness in common-source stages- voltage followers and mu-amps not so much...

[earthtonesaudio]

As typically used in pedal circuits, JFETs exhibit the biggest weakness in common-source stages-

I've often wondered about this... is it because in the common source configuration you have voltage and current gain, whereas in common drain you only have current gain and in common gate you only have voltage gain? 


They do make good switches and diodes though...

[R.G.]

I've often wondered about this... is it because in the common source configuration you have voltage and current gain, whereas in common drain you only have current gain and in common gate you only have voltage gain? 

In common drain (i.e. source follower) circuits, there is 100% negative feedback from the source resistor. This completely hides any gain variations, leaving only the bias variations from Vgs.

In common gate, the current through the device is 100% controlled by the device which feeds the source. Only the voltage gain can vary, and this is dependent on only one parameter, the transconductance of the JFET. The voltage gain is as variable as it would be in Common Source, but the biasing is externally set.

You can combine things to make this better. For instance, you can put a constant current circuit in the source of a common-source JFET setup, and the current source will force the channel current to be what you set up. This requires that the biasing on the gate allow the gate to float to the voltage that lets the JFET do the proper current, but this is less restrictive than trying to make the gate do the biasing all by itself. The drain is forced to the voltage which its resistance times the current says it must be. There are other considerations. By the time you finish it, it's pretty complicated. Once again, it crosses the line where a bipolar or MOSFET would be a more economical choice, all things considered.

[Mark Hammer]

Chopsticks are great for eating many kinds of Asian cuisine.  Would I want to use them for crutches, or to hold up a roof?  Not particularly.  The level of precision and idiosyncratic qualities that JFETs possess CAN make them ideally suited to a particular task, when selected intelligently.  But they aren't the silicon equivalent of 2 x 4 lumber you buy at Home Depot that you can apply to a great many tasks without worry.

[DougH]

IMO, most of the biasing methods a la current sources and other active devices used for propping up a common source JFET circuit end up being like training wheels that are bigger than the bicycle...

[WGTP]

Other than Mu/Syrrp circuits, I haven't had much luck getting Jfets to sound as good as other distortions types.  I ordered some parts the other day and got 10 more Jfets with the hope that they might work better than the ones I have.  I probably selectively used the good ones, until the only ones I had left were bad ones...

To me, this presents an opportunity for DYI and Boutique builders to out perform the large manufacturers who won't/can't take the time to select the good ones for their circuits.   

Thanks Jay for the facts.   

[petemoore]

  I found 'em pretty easy to work with in a single gain stage socket...
  Other than that, the Mu Amp has performance I can live with, the self-biasing makes it a no-brainer, I just socket and swap the actives, and leave it alone when it sounds good.
  I must have 'peripheralled' [built circuits around] 25 Jfets, most with trimpots, but they just didn't muster-up right...
  I did get some cool sounds from multi-Jfet circuits, but kept backing up because something didn't sound quite right, and while re-biasing changes things, whether the signal was ever being treated 'optimally' in each stage caused enough confusion to make other options seem easier to work with.

[Sir H C]

jFETs exist in a lot of otherwise bipolar op-amps because they are easy to graft onto a bipolar process, if it is a CMOS process, they are ignored.

Remember jFETs are not MOSFETs though both have huge variances.  Still you can do some cool things with FETs and doing the EH method of the inverter ICs for the NMOS devices, you can actually get some good matching on devices. 

[drk]

thanks all!
seems i got to "fool with fets" a bit more 

[Zben3129]

You can get some great sounds out of them! Take for example the BSIAB2

Be careful of mixing apples and oranges. A mu-amp is a self biasing circuit that by definition will mask some of the variability issues that JFETs have. As typically used in pedal circuits, JFETs exhibit the biggest weakness in common-source stages- voltage followers and mu-amps not so much...

Well its still a JFET isn't it   
I'd say its more like mixing apples and chocolate covered apples 

Zach

[alanlan]

A tangential moral of all of this is if you see a schematic of a circuit that contains single JFET stages with specific values for the resistors needed, be very, very, very, wary - and in my opinion, expect it to not work at all as drawn - because there is an 80% chance that it won't.

This does not tally with my experience.  I've got a batch of 10 J201's and they all bias to within 5% with fixed resistor values, and they all work.  The devices do have quite a range of IDSS, so much so that I'm 80% confident that any J201 would bias to within 5% or 10% and work.  No other active devices were used in the biasing arrangement.  OK, it does depend on your method of biasing but ways of avoiding trimmers in JFET circuits have been much discussed here.