B1-Turbo on a Chip

[EUVL]

..
..
A while ago, I published an article on alternative follower configurations :

Some other Source Follower COnfigurations

I said alternative, because there have already been the famous Curl / Borbely followers, both in single-ended and the complementary versions. And they have been around for quite a while now. The recent revival of, and the vast interest in the B1-topology, which I find truly amazing, can only be described as a “Pass” phenomenon !!!

Also in my discrete opamp thread from last year, I commented that there could be better ways to do a DIP-8 unity gain buffer than using an opamp :

http://www.diyaudio.com/forums/showt...42#post1480342

i.e. a SMD-B1 on a 10mm x 10mm footprint, using a matched pair of BF862’s.

[EUVL]

You would say, why bother with SMD and DIP-8. It is just as easy with 2SK170s and the like. Or even better, use the 2SK369s for double the transconductance and hence half the output impedance.

Now, imagine that you have a CD player which is currently using an opamp as a unity gain buffer (i.e. output pin directly connected to negative input) to drive, e.g. a Sallen-Key filter. Wouldn’t it be nice to be able to just replace the opamp with a discrete follower, in DIP-8 package, pin compatible to the opamp, and with sufficient low output impedance (say below 20 ohm) to drive any load of such a line-level application.

Ah, but you can still stick 2 TO-92 Toshiba JFETs next to each other, glue them with a dot of thermal adhesive, and they will still fit the footprint of a DIP-8. Also true. Now take a look at the thermal resistance of the silicon substrate in a TO-92 case, and see how much longer it is to get from one substrate to another through two TO-92 cases (5mm of potting compound). Compare that to 2 SOT-23 packages, lying side-by-side, with their top surface glued to a 8mm-thick aluminium heat sink. Bearing in mind that aluminium has almost a factor of 10 better thermal conductivity than potting compound, I bet you won’t need a Ph.D. in thermal physics to figure out which one has a better thermal coupling between the JFET substrates themselves. Why is that so important – differential thermal drift causes DC offset. And you don’t really want that.

[EUVL]

But I see 4 JFETs in the photo! Well, a 2SK170 typically has a transconductance of 30mS, or an effective source resistance of 33 ohm. The 2SK369, as well as the BF862, has almost half that, i.e. about 20 ohm. Take two of those in parallel, and you can get the output impedance down to 10 ohm. The BF862 has the added advantage that it only has half the capacitance of the 2SK170. So even having a pair of them in parallel does not result in any limitations in bandwidth.

How about those SMD resistors ? Two reasons for having the source resistors. Firstly, the Idss of the BF862 is on average too high, at around 16mA. So two in parallel means a DC bias of 32mA, and some CD-player power supplies might find that too hefty. A bit of degeneration help to reduce it to, say, 20mA. You still get plenty of rich, pure Class-A operation. Secondly, they give you some more headroom during the positive swing, to make sure you won’t get to forward biasing the upper JFETs. I also added a gate resistor for each of the signal driving JFETs, just to be on the safe side with any possibility of oscillations between them. You can use any value from 100R to 1k. The capacitance of the JFETs in follower mode is so low that you still get MHz bandwidth with 1k.

[EUVL]

The PCB measures 10mm x 10mm x 0.8mm, with single-sided 35um copper. I have connected the source of the lower JFETs to pin-5, so that a trim resistor can be soldered between pin 4 and pin 5 to adjust DC offset.

I have also tried numerous types of header pins as connections to the outside world, but in the end find that the best solution is a normal DIP-8 IC socket, with 5 pieces of 0.6mm silver-plated copper wire soldered first to the sockets and then to the PCB, at pin locations 3 to 7. You can get really nice, German-made, low-profile IC sockets with fully gold plated pins. They are only 2.5mm thick, compared to 4mm for normal ones. But then they are also a factor of 20 more expensive. There is no free lunch.

So here you are, a B1-Turbo on a chip, and not much more difficult to build than the TO-92 version. How does it sound ? I used a few of them to replace the opamps in the Sallen-Key filter section of my old but still beloved Micromega Stage 1. They just sound great. But of course I am totally biased, as usual.




Patrick


BTW I shall codename this DABF862. I think it should be self-explanatory enough.

[EUVL]

..
(The source resistor values can be varied to suit Idss and bias as you wish; more later.)

[EUVL]

..
The degeneration resistor Rs can be calculated as :

Rs = ( Idss – Id ) / ( Id . Yfs )


With 2x matched BF862 in parallel, and assuming 20mA total bias current,

Idss per FET (mA) ___ Rs (ohm)

11 ___ 1
12 ___ 2.2
13 ___ 3.3
14 ___ 4.7
15 ___ 5.1
16 ___ 6.8
17 ___ 8.2
18 ___ 9.1
19 ___ 10
20 ___ 11

[EUVL]

..
I built a couple of prototypes using matched FETs with Idss around 13mA, not particularly well matched, and used a low-ish Rs of 1.5R (what I have on hand). Here are the measurement results :

[EUVL]

..
Input signal 4V Pk-Pk Sine Wave at 10kHz
Power Supply +/- 9V
DC Offset +2.5mV without trimming and heatsink, stable over 30 minutes


Load Resistor ____ Gain

Open ____ 0.9936
10k _____ 0.9922
2k ______ 0.9858

Effective Zout = 19R at 10kHz

[EUVL]

..
Input signal 4V Pk-Pk Sine Wave
Power Supply +/- 9V
R load 200R

Green Input
Cyan Output

[EUVL]

..
Input signal 4V Pk-Pk Square Wave
Power Supply +/- 9V
R load 10k

Green Input
Cyan Output