Hi. Does anyone know of a well behaved, low cost controller with integrated switch? It will be configured as a non-isolating buck converter running at a duty cycle of 3% most of the time. Switching frequency must be greater than 150kHz since the size of the choke is limited. The supply will be up to 360Vdc. Output power will be about 3W. Thanks.

Ok...Lets say 180kHz... 5.6uSec ...
Conversion ratio that you want 3%...That is 168nSec ON time ???
Keep in mind the DEAD TIME of the controller and leading edge blanking for noise...
Large conversion ratios with short duty cycles don't always work well with traditional BUCK topology.. You are looking for an IC on HV process...
Depending on the application, you can use a "Tapped Inductor" to bring the Duty Cycle up to where the controller can better operate..... The waveform in the tapped inductor has a step between the slopes relative the ratio ....
Since i don't know the intended application, Voltage mode may work better due to the cleaner ramp... If you need Current Mode.. a Valley mode/ emulated ramp may be a better approach.. Average current mode may allow for a smaller inductor depending...
Any consideration for protection against short?? Or are you assuming a open circuit after short ???

Non in the market with integrated switch 3 % duty cycle plus 360 v input
You have only non integrated solution at the present time
High turn ratio for non isolated is also big challenge top inductor my be the solution but still you brew your own controller and related driver

The Power Integrations LinkSwitch-TN might do the job, but it only runs at 66kHz. Inductors aren't particularly large at that power level.

data sheet says 132Khz, sorry if that's incorrect. Also, they will design it for you

You may be right, Bill. I was looking at the LNK302/4/5/6. And you're right about them designing it for you. Their software is quite useful.

PI software is quite evolved. But just make sure you check the waveforms, especially current in the primary switch to make sure there are not any uncontrolled regions, especially turn-on.

If you are going for the tapped buck, be careful - this topology has a RHP zero and can lead to instability. If you stay DCM should be OK.

Years ago, the thought of a buck converter from 360 down to 5V or so was considered to be a bad idea. Failure modes just weren't worth contemplating!

But people do it. Some do it well, some not so well, and some try it then go back to safer isolated solutions.

It seems like nothing is out of the question in today's low cost design world, so you have to be open to all ideas.

Power Integrations parts are probably most evolved in this area, I would agree.

I work mostly in LED lighting but I do know of a few companies that are doing non-isolated controllers with high voltage inputs. Supertex makes a few that can accept a PFC 400V front end stage. Int'l Rectifier has a few controllers that can accept up to 600V for the high side driver. The downside is the duty cycle may not be practical without going with an isolated topology. The Int'l parts I've dealt with are not very stable with a Vo<18V. Plus, some are hysteretic, so there goes the size of the inductor.

Most low power "integrated power switch" devices are usually for flyback typologies. For such low power, non-isolated (no transformer) buck requirement, I'd just use a fixed Ton control (simple timer or one-shot) and variable Toff (frequency) to provide the small duty with plenty of resolution. If Ton is only 0.25uS (just a small inductor needed), at 3% that is still only 120KHz (going 150KHz or above is a EMC compliance problem for online power), ... so decide on frequency after you do some loss equations and define what resolution of control you need. If you want very tiny Ton times, you will likely be pulse skipping mostly. Consider the speed of a comparator needed to monitor switch current (assuming current mode is reasonable). If the magnetics can be slightly more complex, a Sepic would be a better choice, but it increases parts count.

Interesting idea. I had come to the conclusion that a fixed on time was the way to go, but it seems to me that there will be a variable lag in the control loop because the off time will be controlled by the minimum allowable output voltage. In other words it won't be regulating the peak output voltage but the trough. I think a slow loop is inevitable requiring a large output cap. Am I missing something?

Another concern of mine is the volts-per-turn of the choke. The 470uH choke I was considering has 92T which gives over 4V/T. I un-wound the choke and used a 50Hz flash tester to check the breakdown of 10 twists of the wire. I was pleased that it withstood 2kV for a minute then 4kV for a further 2s, but is this a good representation of the quality of the wire's insulation in service? I thought that high frequency can increase the degredation of the wire giving partial discharge even at modest voltage. Does anyone out there have first-hand knowledge of breakdown of commercial shielded cotton-reel type chokes?

You have to look at details of the choke's design too. I have seen off the shelf inductors put in these applications where the start and finish turns cross each other. If you do this, the breakdown will be much lower.

Actually, the naturally-sampled constant on-time modulator has a phase lead relative to the constant-frequency control. Middlebrook showed this long ago, and I made measurements on this in the lab to confirm.

So its not for that reason that you would have a slow loop.

The main problem with the constant on-time with current mode is that you are regulating the off-time, and if you hit DCM there is no signal to use from the current. Hence you just enter a chaotic mode of operation, but probably no worse than the pulse-skipping control.

There wasn't a earlier concern for response time, so actual requirements need to be expressed. Also, Current mode is more problematic for short Ton times. Better to design for voltage mode and use current detection for fault protection. For such small power levels, there may be many post-regulator analog options to accommodate output load response needs. Solving the problem may not be just a single control means. All of it could be easily handled by a very small microcontroller... all you need is a timer.