Improving High Voltage Op Amp's Slew Rate

TI usually have good app notes.
The op amp is directly driving the line?
I'd have thought a driver stage would help, even if it's just another op amp as a buffer.
I googled OP454 IC but it didn't seem to come up (probably just paw trouble)

Ah just found it, spec says typical slew rate 13v/us.
Maybe you need to drive the load push pull?

Yonks since I've messed with opamps but my guess is you are trying to get gain and frequency response (e.g rise /fall time) at the same time which you can't do. Gain x bandwith has a finite limit. So get you gain of 100 in two stages with the last stage being lowish gain and thus faster slew.
I reserve the right for this to be complete tosh, as I'm a bit rusty...but hey ask me a Q about longbows.
Del

Thanks Del

Sorry man, it's OPA454...i built one of the circuits on their application notes.

what do you mean by " the op am is directly driving the line?"

see revised comment...now I've looked at the spec sheet (ignore op amp driving line query, as it obviously is driving the 300ohm load [which I'd assumed was a transmission line, probably erroneously]).
Del

The op amp typical spec is only 13V/μs, so you can't expect too much. Hope you're keeping down any stray capacitance.

Towards the end of this datasheet there's a circuit they claim gives 34V/μs.

Thanks gentlemen. The two stage trick seems to be the most talked about solutions. thnks Del...

John, on what page is that circuit with a 34V/us?

p27

Whoow that's quite an expensive solution there....i popped 1 already. and these things cost more than 60.00 ZAR

I still dont know why i'm getting a rise time of better than 13V/us. must be something i did right(or wrong). But i have to knock off now...(I have a beautiful wife-to be waiting at home)

but i'll try the two stage solution and read up on some pdf's i googled a moment ago.

cheers gentlemen and thanks again

Can you find something pin-compatible, but with a lower spec, and cheap, so you can get the circuit working without trembling every time you switch on?

A compound amplifier circuit can get you eventually the gain and slew rate combination you're looking for but you can quickly have stability problems that may not be possible to solve. If you are wedded to using this specific op-amp though you may never be able to reliably achieve your desired slew rate without going with an H bridge topology that JohnDG suggested.

To answer your question on how to improve the slew rate of an op-amp requires redesigning the op-amp itself. This is not a simple easy to explain collection of tweaks because when an op-amp is slew rate limiting it is operating outside of its linear range. If you wish you can monitor the + and - nodes of the op-amp to see when your amplifier goes non-linear but remember the probing circuitry will effect the circuit performance though.

With the voltage and slew rates you're looking to get, I would consider the Apex Linear Amplifier series of products. Cirrus Logic picked up this series of products awhile back. They're not cheap, but the few times that I used them they worked as stated.

If you are only switching a pulse onto the line why use an op amp? A discrete transistor configuration would likely be a better bet. But if you must, check to be sure you have adequate local current sourcing to the amp (bypass capacitors). If the amp can't get the current it can't source it!

I wonder, do you have cable capacitance involved, which can mean an unexpected pulse driving current, I = C dV/dt?

I suspect your real problem is that you're using an amplifier with internal compensation for unity gain. Externally compensated amplifiers can be "decompensated" for higher minimum gains, such as your G=100, which allows them to be much faster. Some of the APEX amplifiers redfred mentioned have this capability.

There used to be many such parts available. The high-voltage opamp table in our book, AoE, page 213, shows nine of them. One of these parta, TI's 3584, is shown at the right. It can work up to 300V, and if compensated for G=100, the amplifier will have a 200kHz bandwidth and slew at 150V/us.

Although the 3584 is officially "NRND", it's still listed at TI and there are 126 in stock at DigiKey. It is expensive, which might lead one to consider a home-brew approach, along the lines of the circuits we show on pages 168, 256, 369 and 1041.

rcapper asked if you could siomply make a fast high-voltage switch? This is easy to do with mosfets, if you use an appropriate mosfet gate driver. The driver is necessary, because the mosfets have a high gate capacitance that must be charged and discharged.

Favorites of mine are Intersil's HIP4081A and HIP4080A parts. The '4081 works from pulses, whereas the '4080 includes a comparator to work from analog signals. With these you can easily switch as fast as 25ns, or an impressive 2800V/us for a 70V pulse. You can also add gate resistors to slow things down to a reasonable speed. High dV/dt means high dI/dt, and the old equation V = L dI/dt can then rise up and create damaging voltage spikes where you don't expect them, across wiring inductances, so it's a good idea not to go faster than you need.

Here's a block diagram of the HIP4081's operation. Note that is uses two N-channel mosfets. It has a flying capacitor to provide the high-side gate-driver's operating voltage. For example, if your HV is +70 volts, and if you have +12V Vcc to operate the '4081, then when the hi mosfet is on, it'll have +82 volts on its gate and its source, your output, will be at 70V. It will also easily be able to provide a great deal of current to the load, should it be needed.

The mosfet on the bottom is to turn the pulse off rapidly. The '4081 has two such circuits (to drive an H-bridge), but you could simply ignore half of it.

Intersil offers free samples of their chips. They also offer eval boards for $55.

I tried the two stage lowish gain solution and it worked, but my falling edge has a funny ending (looks like an inverted dip) at the end which measures +5V from 0V, but still not affecting the decay time.

There are two issues of concern for an opamp in an amplifier to generate pulses. One is bandwidth, the other is slew rate.

With respect to bandwidth, using the TI 3584 that I mentioned as an example, the datasheet says its gain-bandwidth product (GBW) when compensated for G=100, is 20MHz. This means its -3dB point for a gain=100 amplifier will be 200kHz. The time constant associated with 200kHz is 1 / 2pi 200kHz = 0.8us, or about 2.5us to 90% for a pulse risetime. That's much faster than your 20us goal. Your OPA454 amplifier is unity-gain compensated and has a GBW of 2.5MHz, so I'd expect that in a gain of 100 circuit it'll have a bandwidth of only 25kHz, or a time constant of 40us, and a risetime of about 100us. So you need to use a lower gain.

If instead you set it the OPA454 up with gain=10 (requiring only a 7-volt input), we could expect you'd have a 250kHz bandwidth and a faster than 10us risetime, which should be OK. If you set it up with an even lower gain of say 6 (which will require a 11.7-volt drive for 70-volts out), it should do even better.

The second issue was slew rate. This is often the tough issue for high-voltage amplifiers. Your OPA454 has a spec of 13 V/us, which means you'd expect it to take 5.4us to slew 70 volts. So you'd imagine that it should work in your application. But a parameter rarely mentioned by opamp manufacturers is the relationship between slew rate and input error. It takes a large input error to create the rated slew rate. Bipolar-input opamps require 200 to 300mV, and much more if the they have a degenerated differential amplifier input stage. JFET amplifiers are faster, basically because they have lower transconductance, but they require a much larger input error, several volts.

We discuss this issue in detail in our book, Horowitz and Hill, The Art of Electronics, pages 403 to 410, and show slew-rate, f_T and g_m formulas and measured-data plots.

If the JFET differential-amplifier input stage is degenerated, it will require more than a few volts of input error to achieve full slewing. TI doesn't provide a schematic of their OPA454, but they do provide a schematic for their 3584, a predecessor of the OPA454 (you can get a copy of the datasheet from the DigiKey link I posted), shown on the right. You can see they've got two resistors on the JFET sources, providing input degeneration.

TI/Burr-Brown doesn't say anything about input error and slew rate in either datasheet, so the user is required to do his own experiments. That's basically what you've been doing. When you used the amplifier in a high-gain circuit, you presented it with a small input, which only allowed small input errors, and didn't create fast slewing. If you use the amplifier in a lower-gain circuit, you'll present it with a larger input, which means that while slewing it'll have a larger input error, and will slew faster.

You'll get the best possible results with the largest input, because the ratio of opamp input signal to input error will be larger. That's why I suggested a gain of 6 with 11.7-volt input drive (you can get that drive level from an opamp running on ±15V). Maybe in this way you can barely meet your spec using the OPA454.

Please report back and let us know what you learn.

Thanks again Winfiled,

Your writing is quite fruitful for a young electronics polymath like me.

For my application, I have to drive the input from an FPGA through a buffer which means my input must be somewhere in the range of 3.3V - 5V. This is what i have been simulatting so far.

You seem to be on track to success with a two-stage opamp scheme, with the 1st one operating from +12 or +15V. But if you're simply generating a 70V square-topped pulse, you might consider a MOSFET-switch solution.

You mentioned logic-signal drive. I suggested the HIP4080 family, which is capable of going very fast, much faster than you need. As I mentioned, with a mosfet switch you generally need a gate-driver circuit because the FET gates can have rather high capacitance. But you needn't be as aggressive as the '4081, with its 2.5A gate drive. As an alternate, there's a family of high-voltage drivers that are meant for use with rectified ac-line applications, and as a result they are very cheap and easy-to-use.

I like the parts made by IR, called Half Bridge Drivers. Their IR2302, IR2184 and IR2109 are good possibilities. They all run from 3.3V logic signals. They're somewhat slower, with A 120mA maximum gate drive (less with your gate resistor), and perhaps taking over 500ns to switch. They can handle up to 600V, but they do just fine switching low voltages (even down to 5 or 10 volts).

Newark has the miniDIP version of the IR2109 in stock for $3.31. The IR2304, shown above, has two logic inputs, one for each mosfet switch. But the IR2184 and IR2109 have only one input to select whether the output is high or low, plus a second disable input that opens both switches. There's a cross-conduction-prevention logic to stop shoot-through, and there's a 540ns deadtime delay between switch opening and closings.

What are you working on?