A chemically strengthened window can carry a compressive stress of 800 MPa at the surface and still break when a technician sets it down on a bench. The contradiction is not a manufacturing defect. It is the difference between the strength of the glass and the strength of the part, and the gap between them is almost always at the edge.

What ion exchange actually buys
Strengthening happens in a molten salt bath, typically potassium nitrate at 400 to 450 degrees C. Potassium ions swap places with sodium ions in the glass surface, and because potassium is larger, the swapped region is left in compression. A typical treatment produces a compressive layer 30 to 50 µm deep with a surface stress of 700 to 900 MPa, balanced by tension in the interior.
The compressive layer does one thing: it closes cracks. Glass fails from flaws, and its strength scales inversely with the square root of flaw depth. A surface with a 50 µm crack is an order of magnitude weaker than one with a 5 µm crack. The compressive layer does not let those cracks open, so the part behaves as though the flaws were not there, up to the point where the applied stress exceeds the compression.
The edge is not strengthened the same way
Here is the problem. The ion exchange happens before dicing and edge grinding, or after, depending on the process sequence, and either order leaves the edge in a worse state than the face. Dicing generates microcracks 10 to 100 µm deep. Grinding removes the compressive layer entirely on the cut face, exposing the tensile interior. A chamfer helps by removing the sharp corner where stress concentrates, but a chamfer that is ground rather than polished reintroduces the same flaws at a smaller scale.
The result is a part whose face is 800 MPa strong and whose edge may be 100 MPa strong. Since bending loads concentrate at the edge, that is the number that governs survival.
Three process answers, with different costs
The first is to chamfer and polish the edge after dicing, then re-immerse in the salt bath for a shorter second exchange. Strength recovers substantially, and the cost is one extra process step and the risk of distortion in the second bath.
The second is to etch the edge. A controlled hydrofluoric acid etch of 5 to 20 µm dissolves the crack tips and blunts them, and strength improvements of two to three times are routine. Etching also removes the polishing compound residue that later causes coating adhesion failures. The risk is dimensional change and surface haze if the etch is not uniform.
The third is to design around the edge. Recessing the window into a frame so that the edge is not load-bearing, using a compliant gasket rather than a hard clamp, and keeping fastener holes away from the clear aperture all reduce the stress the edge actually sees. This is free at the design stage and expensive after tooling exists.
What coating does to a strengthened part
Coating adds two effects that are rarely modelled. Deposition at elevated temperature can relax part of the compressive layer: an evaporation process at 300 degrees C for an hour is not the same substrate that went in. Ion-beam sputtering at near-ambient temperature avoids that, which is one reason it is preferred for strengthened parts despite lower throughput.
The second is film stress. A dielectric stack in tensile stress pulls on the surface and adds to any applied load, effectively reducing the compression available to close cracks. A 500 nm stack at 300 MPa tensile is a real contribution on a 1 mm part. The coating design and the strengthening process need to be considered together, and the supplier should be able to say what stress the stack carries.
Testing that reflects handling, not theory
Ring-on-ring biaxial flexure, per ASTM C1499, is the standard for flat glass strength and the right test for a window. Four-point bending works for rectangular parts. Ball drop tests reproduce the handling accident more honestly than either, and are worth running on the first article even though they are not quantitative.
The test that most programmes skip is abrasion followed by strength. A window that survives a pristine ring-on-ring test and then gets wiped with a contaminated cloth in the field is not a 800 MPa part any more. Abrading the surface with a defined grit, then testing, produces a number that predicts field survival far better.
Specifying the edge
A drawing that says "edge polished" tells a supplier almost nothing. What works is a chamfer width and angle with a tolerance, a maximum chip size and the inspection zone in which it applies, a surface finish callout on the chamfer face, and a statement of whether the part will be loaded through the edge or through a compliant seat. Add the strengthening specification: compressive stress, depth of layer, and the test method used to verify both.
Those five lines turn an argument about breakage into an engineering conversation, and they cost nothing to write before the first order rather than after the third.
Product page: strengthened optical windows — specifications, coating options and lead times.