How is your Al2O3 ALD process working? Typically it involves trimethylaluminum (TMA) and H2O. So, I suspect something in your ALD process or masking is going wrong allowing the graphene layer to soak water that is trying to get out upon annealing.

If your deposition of Al2O3 (let's say...) comes from a wet method it is not probable to reach a smooth, bubbleless and crackless film, for a variety of reasons: the water problem explained above by Dr. Gernross could be one of the principal causes, but, if the water problem could be overcome, one would wander which phase of Al2O3 is finally present in your preparation... Typically in those wet chemical methods, the amorphous phase is reached. For crystalline phases one would need sintering temperatures above 800°C (gamma, metastable usually...) and above 1000°C (alpha). These phases probably could be more dense, stable and the surface obtained could be more mechanically robust. Hints to be tried experimentally of course. Greets from Mendoza, Argentine.

@ mr. Lascalea: ALD=atomic layer deposition and it is not a wet method

Following up on the answer of mr. Gerngross: would it not be possible that the water/aluminum oxide etchant goes under the Al2O3 layer after you have etched completely through the layer? Because from the pictures it seems that you get the problem only around the contact pads, where the Al2O3 layer was removed. In that case you might solve the problem by patterning the Al2O3 by dry etching instead of wet etching. Chlorine or bromine containing gases work best for reactive ion etching of Al2O3, but since in this case your resist layer is much thicker than the layer you want to etch any other gas (except O2, which will etch your resist rather quickly) should probably also work (but then the etch rates will be much lower).

Or do you need to keep the graphene at the contact pads? Because dry etching would probably damage it.

As it seems for me from the picture, you firstly do e-beam to make Al-deposition mask? If so, it is important to completely remove resist after lift-off before ALD, otherwise any slight adhesion problems will cause stress cracks and bubbles, especially in case you have wet process (MF-319) with water soaked in ALD Al2O3 structure, which is a good water absorbent.

To exclude graphene interference, try to make the same sample without graphene. But as I see strong ultrathin resist scums (dark brown films) over the pictures, probably you should keep an eye on it.

About dry etching in Cl gases - it would definitely etch Al layer (it is too thin compared to Al2O3). But maybe there could be another possibilities.

Thank you, Rinat. Actually, I deposit Al2O3 layer first on top of  bottom gated GFET and than do lithography ( to etch Al2O3 from contact pads). However, it is possible that residues from previous lithographic steps are still left.

Thank you, Pieter. You are right that cracks appear only near the contacts. However, bubbles appear everywhere. Probably, it a result of wet etching. There is no graphene under that contact pads, graphene is away from that contact pads.

In case of absolute trust in your ALD tool and recipe...

What was the exact step before ALD? If there was any lithography step, it could the reason to have bad adhesion. If not - anyway try to reveal the factors of bad adhesion. For the start I don't think it is a good approach to change the etching step recipe completely, because it seems that adhesion is initially not good for passivation layer. Also maybe it is concerned about too much stress - the film is rather thick (90 nm) for ALD Al2O3 - we always use 10 nm for HEMT passivation - it works perfectly. Actually I don't know whether ALD films are stressfull or not, but why not to try, on dummy samples for example.

By our experience the same cracks and bubbles we have on PMGI containing resist stacks after strong H2O treatments (developing and soaking) when we have deviations off the standard preparation/coating recipes. For insulator films like SiO2 we also had bubbles which grew up during soaking. We associated it with possible contaminations underneath (from technological tools such as O2-plasma in case of VIton contamination over the door).