Parallel to the lamp you can not add any capacitor - it tend to extinguish any discharge...

And regarding the semiconductors on the mains: Mainly with these high power devices that is of no problem at all. Because they have to be rather large to handle the currents, the currents causing the overvoltages are then relatively not that high anymore, so the components can handle them well and so clamp the overvoltages safely.

Moreover all semiconductors there are just thyristors and diodes - the two types which are virtually impossible to destroy (except when overheated; but that is rather difficult given their rating). The sensitive components are all types of the transistors. Not than they will behave much diferent from thyristors on an overvoltage event (got switched ON and then stay like that till the current disappear). But the difference is in the circuit around the thyristors is designed to remove the current externally to switch them OFF (commutation by mains,...), so when it activate itself, the circuit behaves just as if it was normally fired and after some time it deactivates it, so expect a spurious extra pulse on the load, nothing happens. With transistors the circuit expect it will be the transistor itself, to break the current in the circuit, so such uncontrolled activation then means a short circuit with consequent overcurrent and so destruction...

And to break the load circuit is rather impossible at those levels - an arc will form among the broken wires and so keep the path conductive. Only cause other probblems around. When there is current, the arc never extinguish. With "cycling" lamps the cause is not the arc breaking itself, but not being reignited after it get extinct by the zero cross current interruption. If such current interruption happens here, because of no current, there can not be any overvoltage...

And let's not forget that CFLs have loads of these sensitive components(Transistors, Capacitors, etc).
We have really went far in this thread. It started with GE not making CFLs, And we're at switching thyristors.

If th lamp is feeding with DC, i dont see how a capacitor parallel to the lamp would prevent the arc, at any value

In AC case the value matters, but the capacitor does not have to be significant enough to drop the voltage vs the L. All i wanted is to smooth a bit the peak across the Silicon in case of the current interrupting moentarily

I would expect the Silicn to be possible to destroy if :

 - Surge on the line from outside (lightning strike somewhere, ..), not an extreme one but something that Magnetics alone would survive. (That applies if the Silicon is straight on the line and not behind chokes)

 - Interruption in the current, whether from cut off of the supply on the line, or the arc going out (aged lamp, column bumped at Earth level, ...)

The high power devices are capable of standing high currents, but what about breakdown at exceeding allowed reverse voltage (for diode), Vgs or Vds-off (for mosfet), ... I dont expect that the "high power" devices are any better on the voltage ratings too (especially if we want to keep the ON losses down, as i understand those come along with compromise in the max breakdown voltage ratings)

If th lamp is feeding with DC, i dont see how a capacitor parallel to the lamp would prevent the arc, at any value

In AC case the value matters, but the capacitor does not have to be significant enough to drop the voltage vs the L. All i wanted is to smooth a bit the peak across the Silicon in case of the current interrupting moentarily

The setup become unstable, leading to oscillations, which then cause the arc current to get interrupted and so the arc extinct. Of course, some small capacitance is tolerated, but that then won't prevent any overvoltage.
The arc exhibits a negative dynamic resistance, what means once some deviation occurs (e.g. temporarily slightly higher conductivity of the plasma), the voltage becomes held firmly by the capacitor for some short time and during that the ionization level increases too much. The extra ionization level then sustain even after the voltage drops down (it will decay, but that takes time) and in the meantime the capacitor will discharge too much bellow the steady state arc voltage. Now this voltage becomes held there by the capacitor (it needs again a time before it gets recharged back) and during that time the voltage is too low to sustain the discharge, so the arc ceases.
So even when we had constant DC current supply, the capacitor effectively quenches the arc.
This behavior is the main reason, why capacitors are used to actually suppress the discharges, mainly across contacts or so. Include e.g. a fluorescent starter (there the capacitance is too low for the rather slow main tube discharge, but sufficient for the rather fast responding discharge in the starter). Of course, this arc quenching happens when the capacitance exceeds some limit (the arc ionization has to respond faster than the voltage).

 Moreover it would become an extra danger for the semiconductors as well: If the overvoltage still exceeds the breakover voltage of a thyristor, such thyristor just turns ON. But with the capacitor in parallel it means it will discharge it in a very short moment, creating huge current pulse. And that will exhibit huge stress in all of the components, both thermal, as well as electromagnetic forces. Without the capacitors a turned ON thyristor just gets turned OFF by the AC mains and so cause just a spurious elevated current pulse.

I would expect the Silicn to be possible to destroy if :

 - Surge on the line from outside (lightning strike somewhere, ..), not an extreme one but something that Magnetics alone would survive. (That applies if the Silicon is straight on the line and not behind chokes)

 - Interruption in the current, whether from cut off of the supply on the line, or the arc going out (aged lamp, column bumped at Earth level, ...)

The high power devices are capable of standing high currents, but what about breakdown at exceeding allowed reverse voltage (for diode), Vgs or Vds-off (for mosfet), ... I dont expect that the "high power" devices are any better on the voltage ratings too (especially if we want to keep the ON losses down, as i understand those come along with compromise in the max breakdown voltage ratings)

Generally the breakdown alone won't be a problem (except oxide breakdown in MOSFET's and IGBT's, but that can never come from the power terminal side, so from the mains connection).
What is a problem are two things:
- The heat, coming from the related power dissipation. Here if the energy is limited, the devices can handle it without any problems. Because the 10kW XBO ballast components are way bigger than 11W CFL, while the overvoltage events are the same (the atmospheric discharge does not distinguish which load is connected to the hit wire), there is quite large spectrum of the overvoltage energies the XBO ballast can handle without any problem, but instantly kill the 11W CFL.
But in most cases, this mode is not the one responsible for most overvoltage caused electronic failures. Don't get fooled by blown diodes in a fluorescent ballast, they may have fallen as a victim of the other mechanism.

- The response of the device when such overvoltage happens.
Here the overvoltage does not damage the component at all in any way, but it may cause the component to respond in a way, which leads within the circuit to such consequences, which are then deadly for the components. This does not happens on a simple diodes, but it happens on practically all more complex semiconductors. And in nearly all the cases it means the device gets into a conductive state and remains there when the current is flowing.
With thyristor circuits (e.g. a controlled rectifier,...) this is usually of no problem, because the circuit normally operates so, once the thyristor get activated, something else stops the current and that switches it OFF, while the current remains all the time within safe limits. This is the main reason, why thyristors are known as one of the most reliable components in high power electronic - it is just because the way how they operate in a circuit is preventing their destruction even when exposed to occasional overvoltage event.
With the full controlled switching devices (so starting from GTO and all forms of transistors) the circuit usually expects there is nothing else to break the current (in a functional manner), so it is extremely important the device never ever enters the mode, where the normal control electrode does not respond anymore. Otherwise the device remains ON, the current grows way above what all the components can survive, so something breaks as the result of this overcurrent.
Of course, if the circuit would look like it uses to look with thyristors, nothing bad happens either. But the transistors (and GTO's) are used just because for the normal operation they do not need that complex (and slow) commutation circuit around as the thyristors do.
The robustness against entering such uncontrolled mode is usually one of the key component parameters (the SOA'a), even when it does not influence the main operating parameters (losses,...)
And it is not so uncommon, a transistor overvoltage electrical breakdown (so activating the low voltage drop snap back mechanism), yielding to an overcurrent behind a mains rectifier (in a fluorescent ballast), lead to the only destroyed components being the diodes in the rectifier bridge, while the actual transistors are surviving that event (at most with a damage still not yet interfering with the ballast operation).

But still, the higher power devices needs way higher currents, so overvoltage energies, to enter such uncontrollable modes, so you need a less frequent, more energetic event to cause such problem than with a low power design.

Quote from: Ash on February 13, 2016, 09:19:27 AM

If th lamp is feeding with DC, i dont see how a capacitor parallel to the lamp would prevent the arc, at any value

In AC case the value matters, but the capacitor does not have to be significant enough to drop the voltage vs the L. All i wanted is to smooth a bit the peak across the Silicon in case of the current interrupting moentarily

The setup become unstable, leading to oscillations, which then cause the arc current to get interrupted and so the arc extinct. Of course, some small capacitance is tolerated, but that then won't prevent any overvoltage.
The arc exhibits a negative dynamic resistance, what means once some deviation occurs (e.g. temporarily slightly higher conductivity of the plasma), the voltage becomes held firmly by the capacitor for some short time and during that the ionization level increases too much. The extra ionization level then sustain even after the voltage drops down (it will decay, but that takes time) and in the meantime the capacitor will discharge too much bellow the steady state arc voltage. Now this voltage becomes held there by the capacitor (it needs again a time before it gets recharged back) and during that time the voltage is too low to sustain the discharge, so the arc ceases.
So even when we had constant DC current supply, the capacitor effectively quenches the arc.
This behavior is the main reason, why capacitors are used to actually suppress the discharges, mainly across contacts or so. Include e.g. a fluorescent starter (there the capacitance is too low for the rather slow main tube discharge, but sufficient for the rather fast responding discharge in the starter). Of course, this arc quenching happens when the capacitance exceeds some limit (the arc ionization has to respond faster than the voltage).

 Moreover it would become an extra danger for the semiconductors as well: If the overvoltage still exceeds the breakover voltage of a thyristor, such thyristor just turns ON. But with the capacitor in parallel it means it will discharge it in a very short moment, creating huge current pulse. And that will exhibit huge stress in all of the components, both thermal, as well as electromagnetic forces. Without the capacitors a turned ON thyristor just gets turned OFF by the AC mains and so cause just a spurious elevated current pulse.

Quote from: Ash on February 13, 2016, 09:19:27 AM

I would expect the Silicn to be possible to destroy if :

 - Surge on the line from outside (lightning strike somewhere, ..), not an extreme one but something that Magnetics alone would survive. (That applies if the Silicon is straight on the line and not behind chokes)

 - Interruption in the current, whether from cut off of the supply on the line, or the arc going out (aged lamp, column bumped at Earth level, ...)

The high power devices are capable of standing high currents, but what about breakdown at exceeding allowed reverse voltage (for diode), Vgs or Vds-off (for mosfet), ... I dont expect that the "high power" devices are any better on the voltage ratings too (especially if we want to keep the ON losses down, as i understand those come along with compromise in the max breakdown voltage ratings)

Generally the breakdown alone won't be a problem (except oxide breakdown in MOSFET's and IGBT's, but that can never come from the power terminal side, so from the mains connection).
What is a problem are two things:
- The heat, coming from the related power dissipation. Here if the energy is limited, the devices can handle it without any problems. Because the 10kW XBO ballast components are way bigger than 11W CFL, while the overvoltage events are the same (the atmospheric discharge does not distinguish which load is connected to the hit wire), there is quite large spectrum of the overvoltage energies the XBO ballast can handle without any problem, but instantly kill the 11W CFL.
But in most cases, this mode is not the one responsible for most overvoltage caused electronic failures. Don't get fooled by blown diodes in a fluorescent ballast, they may have fallen as a victim of the other mechanism.

- The response of the device when such overvoltage happens.
Here the overvoltage does not damage the component at all in any way, but it may cause the component to respond in a way, which leads within the circuit to such consequences, which are then deadly for the components. This does not happens on a simple diodes, but it happens on practically all more complex semiconductors. And in nearly all the cases it means the device gets into a conductive state and remains there when the current is flowing.
With thyristor circuits (e.g. a controlled rectifier,...) this is usually of no problem, because the circuit normally operates so, once the thyristor get activated, something else stops the current and that switches it OFF, while the current remains all the time within safe limits. This is the main reason, why thyristors are known as one of the most reliable components in high power electronic - it is just because the way how they operate in a circuit is preventing their destruction even when exposed to occasional overvoltage event.
With the full controlled switching devices (so starting from GTO and all forms of transistors) the circuit usually expects there is nothing else to break the current (in a functional manner), so it is extremely important the device never ever enters the mode, where the normal control electrode does not respond anymore. Otherwise the device remains ON, the current grows way above what all the components can survive, so something breaks as the result of this overcurrent.
Of course, if the circuit would look like it uses to look with thyristors, nothing bad happens either. But the transistors (and GTO's) are used just because for the normal operation they do not need that complex (and slow) commutation circuit around as the thyristors do.
The robustness against entering such uncontrolled mode is usually one of the key component parameters (the SOA'a), even when it does not influence the main operating parameters (losses,...)
And it is not so uncommon, a transistor overvoltage electrical breakdown (so activating the low voltage drop snap back mechanism), yielding to an overcurrent behind a mains rectifier (in a fluorescent ballast), lead to the only destroyed components being the diodes in the rectifier bridge, while the actual transistors are surviving that event (at most with a damage still not yet interfering with the ballast operation).

But still, the higher power devices needs way higher currents, so overvoltage energies, to enter such uncontrollable modes, so you need a less frequent, more energetic event to cause such problem than with a low power design.

And it's harder for transistors to switch high currents rather than high voltage.

1. Bigger wires, Bigger contacts, More Silicon, = Easier to break.

2. More current means more heat dissipation because of resistance, And the more components there are, the longer to wires are which means more resistance.

3. Higher current thyristors are more exotic and harder to find = More $$$ = Not a practical idea.

4. This whole problem could be solved by switching to magnetic ballasts, Which are simpler, easier to understand, more reliable, And fewer things to break.

And it's harder for transistors to switch high currents rather than high voltage.

1. Bigger wires, Bigger contacts, More Silicon, = Easier to break.

2. More current means more heat dissipation because of resistance, And the more components there are, the longer to wires are which means more resistance.

3. Higher current thyristors are more exotic and harder to find = More $$$ = Not a practical idea.

4. This whole problem could be solved by switching to magnetic ballasts, Which are simpler, easier to understand, more reliable, And fewer things to break.

That is not true - the chip size vs current dependency is by far not that simple, there are effect, that actually make the low voltage components way more robust. Of course, assuming the margin towards the breakdown voltages is maintained equal in the design (well, in fact it is way easier to have higher margin for lower voltage devices). That actually means for the same power handled, the higher operating, so rated voltage means actually higher losses.
Of course, larger die size brings it's own mechanical problems, but electrically it is way easier to handle higher currents than higher voltages.
It is true with switching devices the main operating modes are ON, so low voltage drop (so quite good dissipation equalization on any component) and OFF state, when there is no current, so no dissipation at all. But the most demanding parts are the transitions, when there is large current and large voltage at the same time, so yielding really high power dissipation. Because an overheat leads to any semiconductor becoming in fact a conductor, the turn ON is usually not that big problem (any potential starting runaway ceases once the device becomes fully ON, even when the temperature had reached a state when the device becomes not controllable on that place), but the turn OFF is the problematic thing (when the thermal runaway causes some part leaking too much, the resulting power dissipation continues to overheat it, even when the device is supposed to be already OFF).
Don't forget we are talking here about small spots few um in size, so the thermal time constants are in us range as well, so the temperature changes way faster than the switching operation. SO even when the device stops working on some spot when turning the complete part ON, before it is turned OFF it cools down and starts to work again. The thing is, the temperatures where the device stops working is way lower than what is needed for it's actual damage or destruction; usually the most heat sensitive parts are not the locally hot ones (well, the complete component is designed so just for that reason).

The main problem starts with the need to use thick layers of low doping levels to reach the breakdown voltage. As the conductivity of the material is proportional to the doping level, it means high series resistance brought into the component, causing high conductive losses. Plus the need for low doping levels means the recombination is there much slower, so the devices operating using minority carriers become way slower, so have higher switching losses, so become limited in switching frequency (and that means higher losses in the magnetic parts)
If you compare multiple components rated at different voltage levels, you will find the ON state resistance (that means not only the MOSFET Ron, but as well the slope of the saturation curve with bipolars - both caused mainly by a thick, low doped Drain/Collector epi) will rise with a steeper than quadratic function of the breakdown voltage for the same silicon area. This is valid above certain critical voltage rating, moreless given for the physical concept (for MOSFET's it is in the range of few 10's of V, for IGBT's the "border" is in 100's of V), below that rating the Ron does not drop that steep with lowering the rating.
But this only defines, how big the silicon should be for the given Ron (or voltage drop) and by itself would not be that big issue.

But there is another problem with higher voltage:
All power devices are practically an array of many small devices in parallel, all on one chip. That means all of them are practically voltage controlled (Vgs or Vbe steers the local drain/collector current density). And because the threshold voltage (the Vgs/Vbe when the current just starts to appear) of all of them has rather negative temperature coefficient. That means if some place becomes hotter, the local threshold voltage goes down, so that means the overdrive (the voltage difference, which then results into the output current density) on that place becomes higher, yielding locally higher current. This leads to more heat emited in the already hotter region, potentially leading to a local thermal runaway.
But against this mechanism goes another one: The transconductance and the finite conductivity of all the diffusions, mainly those on the common terminal (emitter/source). Both of these feature quite strong negative temperature coefficient, so hotter place with the same overdrive has lower current density. This then leads to equalization of the temperature.

These effects are of different relative strength, mainly depending on the actual VCE/VDS: At low voltages, the extra current from the Vt shift is quite small, compare to the current drop from the transconductance/resistance effects, so the current remains evenly distributed over the chip area.
But at high voltages you suffice with way less current for the same extra dissipation, so the Vt shift becomes dominant and so the current tends to focalise and overheat small regions.
To combat this effect on high voltages (at least to cover the usual switching transitions) means extra ballasting resistances in the emitter/source, what means higher Ron, so a need to use even larger silicon. And with really big devices it means higher demand for the uniform temperature, so more sensitive to assembly quality (with low voltage drop operation the higher local temperature usually means the current density drops there, so the device can tolerate worse thermal contact).

There is belief among the electronic community, than the MOSFET's are immune against this effect, in contrast to bipolars, but that is not the case anymore. The devices are designed so, their SOA is able to cover the usual transition time in switching applications, but definitely not in a DC continuous mode. The MOSFET density improvements made the MOSFET's more sensitive (just because of the higher loading per unity area, approaching those common with bipolars), the bipolar design introduced ballasting resistors in the small cells, equalizing the current density even when there are local differences, so make the bipolars way more robust than they used to be in the past (for the same voltage rating).