OPA855: EVB_OPA855_Oscillation problem

Hi Takahashi-san,

Thanks for your documentation for the oscillation debug it is very helpful. 

The main cause of an oscillation like this the distance from In- to the Cpd. That length has some parasitic inductance which resonates with the Cpd to create this 1GHz+ oscillation.

My suggested fix is as follows:

1. Shorten the length to Cpd as you've done. As close as physically possible. How close have you done?
2. The 10ohm for R4 actually helps dampen that resonance. I would keep it there and possibly increase it to 50ohms. 
3. Adding some additional capacitance between the inputs as the R4 resistor isolates the Cpd required to make the device stable. In your case, an additional 100pF in-between the In- and In+ pins would be my suggestion. Placing it between the pads of R2/R1 on the EVM to keep it close to the pins is ideal. Something like this drawing below.
   

Thanks,

Evan

Hi Evan-san,

Thank you for your answer.

Regarding the point 1, it is actually difficult to shorten the distance.

I will try your advice 3. I'll let you know the results.

Takahashi

Hi Evan-san,

Under the initial conditions (Case 1), I added a 100pF capacitor (Cng) between IN- and IN+. While the severe oscillation has subsided, peak oscillation remains around 4.8 GHz.

Is there anything else I can do?

EVB_OPA855_Oscillation problem_2.xlsx

Takahashi

Hi Takahasi-san,

Looks like the oscillation moved from ~3.2GHz to 4.8GHz. The parasitic inductances from that MLCC Cng capacitor could be continuing to pick up a resonance. 

Are you using an actual photodiode for Cpd or are you using a capacitor to simulate a photodiode capacitance? If it is a simulated capacitor, then you could use the ground plane on the top layer or R3 to place the capacitor nearer to the device. 

Otherwise we can lower Cng and Cf to keep the device stable and push out any resonance from any ESL. These would be my recommendations:

• Using a 0201 size capacitor for Cng
• Lower Cf to 1pF and Cng to 10pF
• As low ESL for Cng as possible. I've had success with these Passive Plus 0201N series: https://www.digikey.com/en/products/detail/passive-plus/0201N100FW500-T500/29715992 
• We could increase Rf to better lower Cf and Cng as well. Is there a reason you chose 82ohms? It's on the lower end of what I've seen, and makes the noise gain shaping a little more difficult at these frequencies.
• There are potential fixes in layout as well but we can try to get it stable up with what you have.

These simulations might give you a better idea of the general thought process: 

1. Here the parasitic inductance (1nH) is creating a resonance with the Cpd.
   
   
   
   
2. Adding the 10ohm series resistance dampens that resonance, but the Noise Gain (NG) suffers. 
   
   
3. Adding the capacitor between the inputs helps bring the NG back to intercept the AOL at a good point. 
   
   
   
   

However it's important that this new capacitor doesn't introduce a similar issue with any ESL. You can see the magnitude of ESL that is problematic, so I think my suggestions will help.

Thanks,

Evan

Hi Evan-san,

Thank you for your answer.

I used a 0402 size capacitor for Cng.

And used a capacitor to simulate a photodiode capacitance. But similar to the actual photodiode arrangement, the dummy capacitors are placed with spacing. L occurs, but if possible, I’d like to improve the situation under the current conditions.

I will try 0201 size, following your advice 

And I will try to use lower Cf to 1pF and Cng to 10pF. Is low-frequency stability maintained even if Cf is set to 1 pF?

Below 15MHz, Rf is set to 82Ω to minimize noise gain as much as possible.

I'll let you know the results.

Takahashi

Hi Evan-san,

I tried.

EVB_OPA855_Oscillation problem_3.xlsx

5.Cf : 10pF, Cng : 100pF 

6. Cf : 1pF, Cng : 10pF 

I used 0201 size capacitors for both Cf and Cng. When Cng was set to 100 pF (as in item 5), the oscillation ceased. Switching to the 0201 size seems to have been effective. Thank you for your advice.

However, under condition 6, oscillation is occurring. Do you understand the reason for this?

Takahashi

Hi Evan-san,

Thank you for your answer. I apologize for the late reply.
I understood your explanation.

I have a follow-up question. How is Riso (shown in Figures 10–14 of the datasheet) used? Is its usage different from that of the 10Ω resistor (R4) I used this time?

Takahashi

Hi Takahashi-san,

Good question. On paper the Riso you've shown from the OPA855 datasheet should do the same thing in that it should dampen any L/C resonance and has the benefit of being inside the feedback such that it doesn't create a voltage drop from the APD current. 

However in my experience and in simulation it doesn't work as well to dampen any resonance. You can see here in simulation that there is still some notch in the NG.

Removing the device from the equation could make things more clear. The Riso will isolate resonances from the In- pin but from the output Rf will need to isolate the resonance instead. However it is shorted-out in high frequency because of the Cf.

Note that since VF1 (In-) goes into high-impedance it will overlap with VF2.

Feedback Network with Cf:

Feedback Network without Cf:

Unfortunately you need some Cf to keep the device stable with any input capacitance. 

Thanks,

Evan

Hi Evan-san,

Thank you for your answer. I understand that it doesn't work as well to dampen any resonance. 
Now, I am measuring on my own board. 
Although the situation has improved compared to before the countermeasures were implemented, some slight oscillation is still observed on my own circuit board.
I will consult you after considering the matter, so please wait until next week.

Takahashi

Sounds good. Keep me posted.

-Evan

Hi Evan-san,

I apologize for the late reply.
I am measuring on my own board. Although the situation has improved compared to before the countermeasures were implemented, some oscillation is still observed on my own circuit board. 3.1GHz and 6.2GHz.This is the attached Figure 5-2.
Upon re-checking with the TI EVB, a slight spurious signal is visible around 4.98 GHz.
Are the V+ and V- capacitors involved?
Are there any measures for improvement?
EVB_OPA855_Oscillation problem_4.xlsx

Takahashi

Hi Takahashi-san,

Sorry for the inconvenience, but Evan is traveling for work today and is unable to respond. He will get back to you as soon as he can.

Thank you for your understanding,

Lukas

Hi Takahasi-san,

Apologies for the delay. So the TI-EVB has the spur at ~4.98GHz and your board has a oscillation at 3.1GHz if I understand you correctly. Is the layout of your board considerably different? Is it possible to share your board files with me? You could do it privately with a message if you would like.

I don't think the V+/V- decoupling is an issue. Some quick simulation shows a 3GHz oscillation is a parasitic inductance of about 400pH with your 100pF Cng.

Simulation to sweep L3:

Many off-the-shelf MLCC caps could be in this range. I've had the most success with these 0201N series capacitors from passive plus: https://www.passiveplus.com/wp-content/uploads/2026/04/PPI-0201N-DATA-SHEET-040126A.pdf which are consistently around 100pF of parasitic inductance or lower.

Another option would be to lower the Cf/Cng, but this might require increasing the Rf gain to keep the device stable. What is the exact limitation you have on noise to limit your Rf?

Thanks,

Evan

Hi Evan-san,

I apologize for the late reply.
I'm currently dealing with another issue and haven't been able to work on this one.
I'm sorry, but could you please keep this issue open?


Takahashi

Hi Takahashi-san,

Sure thing. Take your time you can reply to the thread when you are ready. You can also send me a message through the forums.

Thanks,

Evan

Hi Evan-san,

Sorry for the delay.　I was checking it on our circuit board.

Please refer to the "Our Circuit Board" sheet in the attached document.

I've tried various things, but I just can't get rid of the oscillations around 3.1GHz and 6.2GHz.

To reduce the ESL of the 100pF capacitor, six 18pF capacitors were mounted in parallel.

Additionally, to reduce the Q factor, a 1Ω resistor has been added in series with Cng.

I have tried various values ​​for Cf, Cng, Rdump, etc., but I see no improvement.

I tried swapping it for an OPA858 under the same conditions, but it did not result in any improvement.

EVB_OPA855_Oscillation problem_5.xlsx

Could you offer some advice?

Takahashi

Hi Takahasi-san,

Is there are a reason you are staying with a ~100pF Cng capacitor? One benefit from this technique is that Riso will effectively isolate the Cpd, and Cng and Cf can be more optimized on their own. Normally Cpd will determine valid values of Cf, but in this case Cng doesn't need to be that high, both that and Cf can be lowered which will increase bandwidth and help with any resonance for this oscillation. What values have you tried for Cf/Cng?

I'm also not sure if Rdump and Rng are needed. They work in theory but I haven't seen much success with them either. 

I'll try some configurations out in the lab myself this week and see if I can get any success.

Thanks,

Evan