Characterizing and Matching 2n5457 jFET Transistors
There’s a reason why not too many audio DIY folks commit to doing general circuit design using jFET transistors outside of maybe playing with Mu-stages and preamp style clipping circuits.
Many people in the DIY world tend to overlook a basic inescapable fact, that the Common-Source jFET gain stage and other single-ended structures like it require knowledge of Vp (aka Vgs(off) or -Vp) when optimizing for gain and headroom. This is understandable since the design methodology there requires some careful thinking.
Of course, if we’re building something like a “one-to-one” tube preamp iso-emulator architecture, where clipping distortion is the name of the game, this likely won’t be the approach taken. In that case randomly spec’d devices will generally suffice – and there’s the fun of playing a blind-fold game. But for generating clean, high-headroom, preamp stages random-spec on demand this is not the way to go.
Another important case where this applies can be found when jFET’s are used as voltage-controlled-resistors (VCR’s) as seen used in many phasor circuits and AGC/compressors. The Dallas Arbiter “Trem Face” is another example of a jFET based circuit that needs a properly spec’d device to work at its fullest potential.
Indeed, in some circuits there lies a great benefit to using properly and accurately matched jFET devices. In fact, there is a strong potential for extended Phasor range when the matching is done to a higher level of accuracy (ie., at +/- 5mV resolution).
It is important to know that unlike their BJT counterparts, say, jFET’s differ in specs according to a random 2-dimensional variance (see accompanying plot below) // in particular, any method that collapses that relationship into a single number will obviously be wrong by nature. An example of the later approach (blindly borrowed from BJT testing methods, and used by many to this day) will be analyzed and presented on this blog shortly …
To obtain accurate characteristic pair values a multi-tester approach is used, one where the Drain-to-Source voltage is held constant during all measurement phases; a voltage chosen to lie well into the pinch-off region for that device (ie., well above the maximum listed value for |Vp|). For my lots of 2n5457 devices I chose a test voltage of 7.50 volts, as |Vp|max = 6 volts.
Three pairs of data points are extracted per device, one for Idss (at Vgs=0), one near cut-off (at Id=10uA) and a mid-point value (somewhere between Idss/3 and Idss/2) … from this 3-point data a good estimate for Vgs(off) can be derived … leaving us with a numerical characterization pair (Idss@7.50v and Vp) for each device … if all devices are tested at the same reference voltage then these two numbers can be used to establish a strong basis for matching. Assuming matched characteristics throughout the hole transfer range of two devices for which Vgs(off) and Idss@7.50 values coincide closely.
IN case you’re wondering, Temperature does not factor in jFET testing as much as it does with BJT’s. I’m not sure, but I think this is because the forward-biased diode junction in a BJT is strongly dependent on temperature, firstly, whereas in a jFET this junction is reversed biased and so temperature plays a secondary role in the device’s behavior … indeed, touching the jFET with fingers during testing does not cause a strong migration of readings like it does with BJT’s, and so the testing can be done reliably without requiring a temperature controlled forced-air system.
Below, we can see what my three un-related lots of 2n5457’s look like all graphed out once characterized … notice, all values are found lying well within the rated limits for the 2n5457 device: 1mA < Idss < 5mA and -6v < Vp < -0.5v (see datasheet)
The following two pdf files demonstrates the methodology and provides tabulated results. In a total of 250 or so devices, we find very few “perfect” matches if we use +/- 5mV as a benchmark … at +/-10mV (using adjacent Vp values) the choice gets a little better, but still this study shows that even with so many devices the likely-hood of good matching is actually quite poor considering the spread of potential values / not a surprise, and something that “adds” to the jFET’s ill reputation
if researchers were to characterize their 2n5457 devices, say, using the same technique and parameters then the possibility of trading devices for obtaining matched sets becomes a reality
As far as sets go, we find the presence of 9 closely-matched QUADS in the following range: -1.49v < Vgs(off) <-1.38v of the data ... and of course, many more triplets (underlined) or pairs ... that is, those devices for which Vgs(off) is an identical estimated number and Idss@(-7.50v) are the closest to east other - ie., well within 10% from each other ...