Forums

SPICE gets it wrong

Started by Tim Williams May 7, 2017
"John Larkin" <jjlarkin@highland_snip_technology.com> wrote in message 
news:vkp1hc1a8ijpfnrpbvo9gei9re09ise5lm@4ax.com...
> Probably not, although they might contribute at high currents. Didn't > the sim need nanohenry inductances to get it to oscillate? The base > lead of even a big transistor can have a skinny wirebond.
That's what I meant about "physically realistic" -- wire bonds are a couple nH, and even with the loose-wired layout, I can't blame more than about 20nH on the circuit itself. More rather, it shows that the simulation is pretty close to instability. We're not talking much phase shift at this frequency, just a little is enough. Bitrex's hybrid-pi approximation also shows a peak at a modest frequency, though it doesn't seem quite close enough to outright oscillation. For both models, a little delay or phase shift would seem to do the job. Tim -- Seven Transistor Labs, LLC Electrical Engineering Consultation and Contract Design Website: http://seventransistorlabs.com
On 05/08/2017 08:07 PM, Tim Williams wrote:

> Bitrex's hybrid-pi approximation also shows a peak at a modest > frequency, though it doesn't seem quite close enough to outright > oscillation. For both models, a little delay or phase shift would seem > to do the job. > > Tim
The Z_o of the 3904 into the C_pi of the TIP31 plus the 3904's current gain is probably enough to push it over. It's oscillating as a unit, right? The collector of Q1/Q4 feeding back to the base of Q2/Q3 thru the diode capacitance or power supply leads or something?
"bitrex" <bitrex@de.lete.earthlink.net> wrote in message 
news:jl8QA.123015$uG1.107099@fx08.iad...
> The Z_o of the 3904 into the C_pi of the TIP31 plus the 3904's current > gain is probably enough to push it over. It's oscillating as a unit, > right? The collector of Q1/Q4 feeding back to the base of Q2/Q3 thru the > diode capacitance or power supply leads or something?
It's certainly something about the 3904 having some delay or phase shift or funny impedances, and wrapping that around another transistor. After some more playing around, I've concluded that: - Q3 collector circuit doesn't play much role. - An R+C from Q3 base to GND helps a lot. In some cases, it can fully stabilize the circuit (MJ15028 and 2SC4821 use around 100R + 470pF). - I seem to have found unconditional stability: an R1 || L || (R2 + C) between Q3 emitter and Q4 base. Values are R1 = 220R, L = 8.7uH, R2 = 22R, C = 220pF. This, and 220R + 220pF from Q3 base to GND, seems to work on every transistor I try! I've taken over Q3 collector as a "sense" for what's going on, since Q3 current is modulated heavily when oscillating, meanwhile all the other node voltage changes are fairly modest. I'm using 10uH || 470R there (between Q3 collector and VCC). Tim -- Seven Transistor Labs, LLC Electrical Engineering Consultation and Contract Design Website: http://seventransistorlabs.com
In related news, I've discovered this circuit drives Darlington transistors 
well beyond ludicrous speed: we're into plaid speed!

I have a 2SD560 (NEC, probably 80s or 90s) which claims 1/1.2us (rise/fall) 
at hFE(sat) = 1000.

In comparison, the fastest transistor (2SC4821) does:
turn-off delay 125ns
rise time 30ns
turn-on delay 614ns*
fall time 59ns

(*A consequence of the Q3 base damper, see last response to bitrex.)

The 2SD560 does:
turn-off delay 160ns
rise time 160ns
turn-on delay 750ns
fall time 230ns

Tim

-- 
Seven Transistor Labs, LLC
Electrical Engineering Consultation and Contract Design
Website: http://seventransistorlabs.com


"Tim Williams" <tiwill@seventransistorlabs.com> wrote in message 
news:oeod3g$id1$1@dont-email.me...
> Speaking of SPICE, here's a nice little circuit: > https://www.seventransistorlabs.com/Images/BakerDarlington.png >
On a sunny day (Mon, 8 May 2017 16:25:09 -0500) it happened "Tim Williams"
<tiwill@seventransistorlabs.com> wrote in <oeqniq$mf8$1@dont-email.me>:

>"John Larkin" <jjlarkin@highland_snip_technology.com> wrote in message >news:vhf1hcltlt23aqpts4stfqq50eb5hb41q5@4ax.com... >>>Interelectrode capacitances aren't enough to make it pop off? You need >>>the lead inductances? >> >> I think that wirebond inductances are the big contributor to >> transistor oscillation. >> > >At 5MHz? > >With a TIP31C? > >Tim
From memory, the discussion at that time was that Texas Instruments uses a different fabrication process than Motorola (MJE3055), but that is from hear-say. Have not used TIP<xx> ever since then though, why take chances.
On a sunny day (Sun, 07 May 2017 19:59:37 -0700) it happened John Larkin
<jjlarkin@highlandtechnology.com> wrote in
<vmnvgcpmkl85uf7gb0ub1k680f2hlot825@4ax.com>:

>On Sun, 7 May 2017 19:14:26 -0500, "Tim Williams" ><tiwill@seventransistorlabs.com> wrote: > >>Speaking of SPICE, here's a nice little circuit: >>https://www.seventransistorlabs.com/Images/BakerDarlington.png >> >>I discovered the oscillator on the left, when attempting to build a >>Darlington transistor pair with a Baker clamp. (In practice, R10 comes from >>a logic voltage input, and R2 is a load, often inductive. R14 is optional, >>but exacerbates the oscillation, making it reproducible.) >> >>The model on the left, of course, doesn't produce any oscillation: it >>reaches equilibrium within a couple hundred nanoseconds. (Which is kind of >>suspiciously fast for a TIP31 to do much of anything, I might add.) >> >>The simulation on the right, showing semi-reasonable parasitics, was >>necessary to reproduce the oscillations. The parasitics are in reasonable >>locations, but their values are completely unreasonable. >> >>The left circuit appears to be a limit cycle of an chaotic system; adding C1 >>to the real circuit introduces what looks like rising-edge crossover >>distortion, and adding a few other things causes all-out chaos. >> >>(Incidentally, if you'd like to volunteer solutions that stabilize the right >>hand circuit -- other than reducing and removing the parasitics, which >>apparently isn't an option -- I'd be interested to hear them.) > >Use a mosfet?
Indeed. Or drive it from a current source.
On 08/05/2017 01:14, Tim Williams wrote:
> Speaking of SPICE, here's a nice little circuit: > https://www.seventransistorlabs.com/Images/BakerDarlington.png > > I discovered the oscillator on the left, when attempting to build a > Darlington transistor pair with a Baker clamp. (In practice, R10 comes > from a logic voltage input, and R2 is a load, often inductive. R14 is > optional, but exacerbates the oscillation, making it reproducible.) > > The model on the left, of course, doesn't produce any oscillation: it > reaches equilibrium within a couple hundred nanoseconds. (Which is kind > of suspiciously fast for a TIP31 to do much of anything, I might add.) > > The simulation on the right, showing semi-reasonable parasitics, was > necessary to reproduce the oscillations. The parasitics are in > reasonable locations, but their values are completely unreasonable. > > The left circuit appears to be a limit cycle of an chaotic system; > adding C1 to the real circuit introduces what looks like rising-edge > crossover distortion, and adding a few other things causes all-out chaos. > > (Incidentally, if you'd like to volunteer solutions that stabilize the > right hand circuit -- other than reducing and removing the parasitics, > which apparently isn't an option -- I'd be interested to hear them.) > > I think the TL;DR is: > 1. TIP31Cs suck as switching transistors to begin with, > 2. The TIP31C SPICE model sucks, despite them having half a century to > try and get it right. > > Tim >
Why do you call it Darlington? I thought Darlington meant collector to collector. The second transistor in a Darlington is already saturation clamped - only the first transistor could be clamped. Without your clamp diode there is unlimited base current in the second transistor - perhaps that is upsetting things? Insert resistance in the collector of the first transistor? Or wire it collector-collector as a true Darlington. piglet
"Jan Panteltje" <pNa0nStpealmtje@yahoo.com> wrote in message 
news:oerptb$1lo1$1@gioia.aioe.org...
> From memory, the discussion at that time was that Texas Instruments > uses a different fabrication process than Motorola (MJE3055), > but that is from hear-say. > Have not used TIP<xx> ever since then though, why take chances.
Wouldn't be surprised; 3055s were /terrifyingly/ bad. By the 80s, most people were selling "okay" dies as 3055s. Dies that would've been used in MJ15020s and such. I have a few "2N3055"s, dated late 80s, that are downright reasonable in switching speed and Vce(sat). AFAIK, the same is true of TIPs, and pretty much every other early crap part that survived multiple generations of fabs. I do have a 2SD880 that seems to perform even worse than the 2SD401. Very slow. I suspect it's closer to the original TIP31, or 2N3055 for that matter. Tim -- Seven Transistor Labs, LLC Electrical Engineering Consultation and Contract Design Website: http://seventransistorlabs.com
"piglet" <erichpwagner@hotmail.com> wrote in message 
news:oervvf$qco$1@dont-email.me...
> Why do you call it Darlington? I thought Darlington meant collector to > collector. The second transistor in a Darlington is already saturation > clamped - only the first transistor could be clamped. >
The emitter follower action is the first feature of the configuration that comes to mind, but it's true the collector clamping action is also inherent, so it's a misnomer to call it that.
> Without your clamp diode there is unlimited base current in the second > transistor - perhaps that is upsetting things? Insert resistance in the > collector of the first transistor? Or wire it collector-collector as a > true Darlington.
The "unlimited" current seems to help more than it hurts, while the clamp is doing its job -- at turn-on, it's goosed with enough extra charge to swing the output node down /and/ stay fully saturated for a few hundred nanoseconds. Here's the full circuit, adjustments included: https://www.seventransistorlabs.com/Images/BakerDarlingtonSch.png Ironically, this simulation oscillates during the /off/ period! Dunno if it's parametric, or something accidentally biased, or what. Putting 300 ohms across L1 seems to fix it. BTW, the ferrite bead L2 is a reasonable model (though a linear one), and C4 has a tinge of ESR and ESL. No supply R or L, and no other stray L beyond what's shown. (L3 really should have a model, since it's on a somewhat lossy ferrite core.) Tim -- Seven Transistor Labs, LLC Electrical Engineering Consultation and Contract Design Website: http://seventransistorlabs.com
On 05/09/2017 05:56 AM, Tim Williams wrote:
> "Jan Panteltje" <pNa0nStpealmtje@yahoo.com> wrote in message > news:oerptb$1lo1$1@gioia.aioe.org... >> From memory, the discussion at that time was that Texas Instruments >> uses a different fabrication process than Motorola (MJE3055), >> but that is from hear-say. >> Have not used TIP<xx> ever since then though, why take chances. > > Wouldn't be surprised; 3055s were /terrifyingly/ bad. > > By the 80s, most people were selling "okay" dies as 3055s. Dies that > would've been used in MJ15020s and such. I have a few "2N3055"s, dated > late 80s, that are downright reasonable in switching speed and Vce(sat).
The Radio Shack near me went out of business (again) and I bought up all the bagged 2N3055s they had sitting in their bin for about 2 bucks for 10. I should test 'em to see how good (or bad) they are. PS: need any twist-on BNC connectors? Nickel-plated brass with gold contacts. I get you good price. Oh wait I already asked ;-)