I am looking for an optocoupler that can sustain a 400kHz signal. Now the question is how can I find out from a datasheet whether it does?

Looking at a typical PC817 datasheet I see the following response times

  • rise (4-18µs)
  • fall (3-18µs)

and a cut-off frequency of 80kHz.

Does that mean a PC817 can be used up to 80kHz?

What rise/fall times do I need for 400kHz? (around 300ns?)

Or is there another factor that affects the viability of the optocoupler?

Can you recommend some common and fast enough optocouplers for this speed?

  • \$\begingroup\$ What do you need for specs? fmax, jitter, duty cycle? tolerances. They do make MHz range optos \$\endgroup\$ May 20 at 0:39
  • 1
    \$\begingroup\$ I’m lazy, so I choose a product that has internal processing. Something like the Sharp PC900 and there’s many other similar devices. This might sharpen up your search criteria. \$\endgroup\$
    – Kartman
    May 20 at 1:46
  • \$\begingroup\$ iCouplers are transformer based and can get a lot faster than that but requires a logic power supply on both sides. \$\endgroup\$
    – DKNguyen
    May 20 at 1:55
  • \$\begingroup\$ There are optocouplers based on phototransistors (slower) and photodiodes (faster). \$\endgroup\$ May 20 at 1:57
  • \$\begingroup\$ Is it a digital signal or something else? Digital isolators might be an alternative. \$\endgroup\$
    – Lundin
    May 20 at 6:42

5 Answers 5


You should read datasheets with a bit of a jaundiced 1 eye.

The "typical" response times are valid iff you have a load resistor of 100 ohms. That is rarely true in a typical application since the voltage you could reliably expect to see across a 100 ohm load would be rather small.

If you want to have (say) a 4.7k ohm load resistor, then the rise and fall times will be much longer, more like 50 to 70μs (typical, at 25°C, and without saturating the transistor).

enter image description here

For high speed applications it's better to use a logic-output optoisolator such as 6N137. It uses a photodiode internally on a chip that delivers a logic output and has guaranteed fast behavior under useful conditions.

  1. To view someone or something in a negative or uncharitable light that has been informed by past experiences.
  • \$\begingroup\$ This is great as this explains the variance of the response times. But can I calculate a frequency range straight from the response time graph? Or are there other factors in the mix? \$\endgroup\$
    – tcurdt
    May 20 at 8:14
  • \$\begingroup\$ The transistor PT or diode PD output capacitance and conductance controls the slew rate. Thus keep bias ON in the low level current. Dual PD's with feedback guarantee that in those with 1 emitter and dual outputs. Otherwise use the worst case CTR's and compute bias accordingly for max dV/dt=Ic/C \$\endgroup\$ May 20 at 15:13
  • 2
    \$\begingroup\$ Frequency range depends on what kind of waveform you are expecting at the output. Saturation (which you typically get in a switching situation) will also slow the switching. Easiest way is to get an idea from the graph then simulate and add a safety factor, or just buy a properly specified part if you're anywhere near the limits. PC817 is more suited for isolating PLC inputs and that sort of hundreds of microseconds to millisecond response isolation. For 400kHz, forget about using the PC817. \$\endgroup\$ May 20 at 15:16
  • \$\begingroup\$ Rather than jaundice \$^1\$, I suggest view with focusing on all test conditions then use 1st principles then verify with simulation for C out and CTR min ~ max and compute worse case at Tmax. Plots are always typical and tables = guarantees for conditions given only. \$\endgroup\$ May 20 at 15:46
  • \$\begingroup\$ @SpehroPefhany The PC817 was just an example. I kind of expected it to be too slow for a digital output isolation :) But I couldn't really figure that out from datasheet. Hence my question. The signal will be steps for a stepper driver. So a digital signal. \$\endgroup\$
    – tcurdt
    May 20 at 18:03

There are some tricks to getting faster performance from crummy phototransistor isolators.

The simplest is a B-E resistor -- when the base pin is available (e.g. 4N35). This greatly reduces CTR, especially at low currents, but provides a sink path for stored base charge.

Basically, the C-B junction is photoactive; the photocurrent then flows into the B-E junction, amplifying it by hFE. The B-E resistor shunts some of that current, so the transistor stays off until Vbe is reached (hence, CTR drops precipitously below operating current), and turning it off faster (discharging Cbe). This allows rise/fall times around 2µs.

We can go even further with something like:

4N35 transimpedance amplifier

This has bandwidth similar to what you're asking, actually. It's a transimpedance amplifier, using the phototransistor as combined input current source and input transistor. It can probably be improved further (I forget now exactly what the cutoff frequency was, but it had a 1st order response, implying it could be compensated to go much faster).

Not that I would encourage using such circuitry -- this is more of a gimmick, interesting that it works at all, but not very practical.

Ready-made devices exist, with better optimized properties -- 6N136 is a photodiode paired with a general-purpose NPN, greatly saving on capacitance and stored charge. You can even add a B-E resistor to speed it up (though, its CTR is quite low to begin with) -- but therein lies the problem, as the base pin itself is extremely sensitive to injected charge: the CMRR of this device is appalling. There is, however, an improved version: SFH6345 omits the base pin, and has improved shielding between emitter and photodiode.

This is also at the point where logic-level devices, like 6N137 and others (HCPLxxx, etc.) show greater promise, as well as entirely non-opto devices like TI's capacitive isolators (ISOxxxx), and the monolithic transformer based devices from Analog Devices (ADuMxxxx), Si Labs, Infineon, etc. These are MUCH faster (>100Mb available) and offer excellent CMRR.

Do note that most of these are entirely digital methods; if your signals are analog, you may be better off digitizing them first. There are matched-photodiode isolators (IL300, HCNR201, etc.) which still need some offset/gain calibration (just, much less than a single photodiode or phototransistor type, and which are stable vs. aging), or there are AC modulation methods which may be of interest (say, AM or FM into a transformer).

  • \$\begingroup\$ Good ideas for discrete telemetry . These are just step pulses for a step-rate controlled stepper motor. with the smallest pulses possible thus duty cycle is reduced with higher If and keep well below 50mW \$\endgroup\$ May 20 at 16:33
  • \$\begingroup\$ Yes, thanks for posting how to (painfully) get rated speed from a sluggish phototransistor. We used the base resistors on the 4N35s back in the day. Ofc PC817 omits that pin for various reasons. A 75ns (max) 6N137 is about 20 cents in 100's so I think that may be a better solution up to a few Mbaud. There's also a Motorola part number that I can't remember atm. \$\endgroup\$ May 20 at 16:45

If you search at Mouser or Avnet rather than Digikey you will find more high-speed optoisolators in the 1 to 10MHz range that are low cost.

6N137's are the best choice for this app. by far. Multiple makers and distributors and low cost, good supply.


6N137 – High Speed 10MBd Optocouplers

You may consider others similar eg. 6N135 and keep biased on at low currents for 1Mbd rates with low R loads. For faster RC=T results. e.g. https://www.mouser.ca/datasheet/2/239/lite_s_a0003567639_1-2272018.pdf High risetimes may be reduced this way, rather than the full off toggle rates.

This way keeps the effective impedance from going high with Vce/Ic at say 5% IMAX.

e.g. 6N135-500E lots in stock but long LT. $2.66 1 Mb/s https://www.mouser.ca/datasheet/2/678/AVGO_S_A0003869335_1-2574502.pdf

There are many key design tricks in using any diode or photodiode for max. data rate.

  • avoid Off-state currents where capacitance is maximum at 0V and incremental resistance is highest result in in large RC=T risetimes and reverse recovery times
  • keep bias current ON but maximize the Max/Min ratio for best SNR
  • Choose the worst case or lowest CTR and consider PD's vs PT outputs.
  • PD's have very stable fixed CTR (0.5mA/mW) but LED's have a wide tolerance on ESR or Rs or bulk resistance that affects Vf at max current and also efficacy at any current which controls the CTR ratios in LED-PD output outputs. Phototransistor (PT) outputs add considerably more variation due to hFE tolerances of > 300% yet offer greater maximum sensitivity.

The problem with optocouplers is that they are two analog devices (an LED and a phototransistor) and the actual delay will depend on the driver and load. So asking for the max frequency brings you the question of 'what is it connected to?'

If what you need to send is a digital signal and one of the requirements is the frequency or data rate, you may be better off by choosing a digital optocoupler, where the output is digital with a fairly sharp edge and most importantly quite independent of the load.

For example, the H11L1M is rated for 1 MHz, you will have no problems at 400 kHz.

  • \$\begingroup\$ Interesting. I didn't even know such digital optocoupler exist. \$\endgroup\$
    – tcurdt
    May 20 at 17:58

Mouser has a separate category for high speed optocouplers: https://www.mouser.co.uk/c/optoelectronics/optocouplers-photocouplers/high-speed-optocouplers/

You can look at the data rate column. For example if an optocoupler is marketed for 1 Mbps NRZ it will likely support a 1 MHz square wave.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.