PCB fab characterization - undercut in DorkbotPDX batch order

Most of the time people making PCB simply design the board exactly as they want it to look, send it off to the fab, and trust that they'll get back a board that matches what's on the gerbers. Unfortunately this is not actually the case! The gerbers are used to create a photomask, which is then exposed onto a photoresist-covered panel and etched.

Most PCBs are patterned using a ferric chloride or copper chloride based etch which is almost perfectly isotropic - it etches equally in all directions. This results in the mask being undercut by approximately the thickness of the copper layer (35 μm for 1oz copper).

PCB cross section showing sloped sidewalls from undercut (homemade board)

Ever since I did my first BGA board and noticed some of the pads looked a bit small I've wanted to do a formal study of mask undercut on this process so that I can design future boards with this in mind.

While the undercut is small and insignificant for many applications, it can become a significant issue when working with fine-pitch BGAs, 0201 sized passives, and other very small components!

This was my first time imaging PCBs at high magnifications (more than the 30x of my inspection microscope) so I fooled around a little bit with techniques. Somewhat surprisingly I found that darkfield illumination gave better results than brightfield - brightfield images tended to be far too bright due to reflections from the soldermask.

My first test was one of the leftover boards from my BGA test in January. In darkfield the pads stood out very well and were easy to measure to within a few μm. All dimensions were taken with the focal plane at the top of the trace. At some point in the future I plan to cross-section a board and measure edge profiles.

200 μm nominal diameter BGA pads (darkfield image). The circles below and to the left of the pads are damage to the soldermask from a misaligned BGA reflow attempt.

The pads are on 500 μm centers and are nominally 200 μm diameter with 300 μm space between them. A quick inspection shows that the pad size is much smaller than the space! Actual measurements show that the left-hand pad is 85 μm in diameter and the right is 67 μm, for undercut of 115 μm and 133 μm respectively!

Further inspection showed that one of the isolated pads (not connected to a trace) had been etched away entirely.

Brightfield image showing empty hole in soldermask where a pad was etched away. Note strong glare from soldermask in brightfield mode.

It is also interesting to note that the soldermask apertures are almost exactly 200 μm in diameter despite the gerbers including some clearance around the pad.

A higher magnification image allowed me to measure trace and pad sizes more precisely.

Higher magnification image of same area

The 200 μm nominal pad is actually 89.5 μm in diameter and the 152 μm trace is actually 57.9 μm. The undercut was over 50% in this case - 110.5 μm on the pad and 94.1 μm on the trace.

In order to verify that the undercut was not specific to this one board I tried a more recent one - my 0201 passive test panel from last week.

0201 capacitor footprint

Both pads are nominally 250 μm wide. Actual measured dimensions are 187 μm and 190 μm for undercut of 63 and 60 μm respectively.

Higher magnification view of a single pad

This pad measured 189 μm wide. Edge roughness was very small and hard to measure but it looks to be under 5 μm.

The same board also contained an 0.8mm BGA footprint. Pads are nominally 400 μm on 800 μm centers.

0.8mm BGA footprint

The results were consistent with previous measurements - actual pad sizes are (clockwise from top left) 329 μm, 329 μm, 331 μm, and 332 μm. Undercut values are 71 μm, 71 μm, 69 μm, and 68 μm.

In conclusion, the fabricator used by the DorkbotPDX batch order exhibits nontrivial undercut on small features. Observed undercut values ranged from 68 to 133 μm and were fairly consistent within a single board but varied from board to board. The 0.5mm BGA test board exhibited an average undercut of 113.5 ± 19.5 μm and the 0201 test board exhibited an average undercut of 67 ± 4 μm.

Multiple Lithography in Homemade PCBs

Earlier this week I decided it was about time to try making a board for some of the 24AA16 EEPROMs I had sampled from Microchip a year ago. It's a 16kbit I2C EEPROM with five pins: power, ground, I2C data and clock, and write protect.

Five pin package - piece of cake, right? But, just to add to the fun, the package I picked was CSBGA with balls about 250μm apart!

From the packaging specification, it can be seen that the balls are 150 μm diameter and spaced in a 2x2 grid 570μm x 520μm with one more in the center. This is a little smaller than my laser-printer contact lithography process can comfortably resolve. What to do?

Conveniently I have a metallurgical microscope that I've managed to coax into service as a projection lithography system. The field of view is, however, far too small to do an entire PCB.

After a little thinking I decided to try a multiple-exposure technique. The first step was to design a board layout in ExpressPCB (my preferred CAD tool is kicad but Express is a little easier for super simple layouts) with a 4-pin SIL header going to a rectangle of copper a little bigger than the CSP footprint. I also made a second mask containing the BGA footprint and tracks going out to four large pads, and printed it at 4x actual size. The center ball is WP# so I tied it to Vdd rather than breaking out to a separate pin.

Mask design

I then printed a mask on my printer, exposed onto precoated PCB, developed (1% w/v NaOH in distilled water), and etched (6 parts 3% H2O2 : 1 part 32% HCl at low heat) as with my standard PCB process.

The next step was to strip the existing photoresist since it had been exposed to light during the etch process. A few drops of acetone did the trick nicely.

I then spin-coated the board with fresh photoresist, using my standard mixture (Shipley SP24 photoresist diluted 50% v/v with acetone for a thinner layer), soft baked on a hot plate, and exposed the BGA mask onto the copper rectangle. After developing, this was the result:

Second photomask on top of etched metal1

Closer view showing edge quality

Not surprisingly, the resolution and edge roughness were vastly better than the contact lithography process. (I've scaled the same technique to 20 μm half-pitch on silicon and there's room to go a lot further.) In retrospect the traces were a little too small considering that the copper layer they're sitting on is 35 μm thick, but this was a mask design error and not a process issue.

Since the thin photoresist I use is harder to see on copper than the thick stuff the board came coated with, I tossed it in the etchant for a couple of seconds to make it more obvious what was being masked.

After a couple seconds in the etch bath

The copper pad was also a bit larger than it needed to be and exceeded the FOV of the lithography system (note the unwanted photoresist shorting the pads together). I gently scraped this away with a #11 scalpel blade under 30x magnification and briefly etched to confirm good separation.

Surgery time!

After etching, no shorts

I then etched for a couple of minutes and removed the board to see how it was doing.

Almost done etching

Things still looked very good, all traces were intact. For traces with such a high aspect ratio (about 40 μm wide in 35 μm thick copper) things looked surprisingly good, but it needed a little more time.

Overetched. (photoresist stripped before taking this pic)

Unfortunately I overetched, one of the traces was gone entirely and another was seriously damaged. Some residue was still in place between two of the pads. Perhaps better agitation would help?

In either case, had the traces been a little larger (perhaps 75 μm) or the copper a little thinner it would have worked beautifully. The lithography itself was flawless and even though the board was not usable it appears the technique is feasible. Given a mask respin this same board could be fabricated with little difficulty.