packages
I wonder why octagonal IC packages in the LQFP times have never been a thing.
In my opinion they would allow easier pcb routing, they take less space and weight for the same die size, shorter leads to the die compared to square packages as well.
Pick and place machines wouldn't have a problem with them, and EDA software is fine with 45° routing anyway, so it seems like a logical thing to do.
Yet I've never seen one.
Any idea why?
This picture can help to see the difference between BGA and WLP
Smaller packages, more specifically packages with less pins, are usually less expensive. Usually, because it also depends on the technology; QFP technology for instance is cheaper than CSP (Chip Scale Package). I presume this WLP (Wafer Level Package) for the LPC1102UK
is the smallest ARM package to date, body is 2.17 x 2.32 x 0.6 mm, with 16 bumps. That's damn small, yet it costs almost USD 5.00 quantity one (Digi-Key). Even at 3000 pieces the price is still over USD 2.00. (Remember, this is a Cortex M0, the lowest end ARM.)
From recent limited research I found that there are few Cortex M devices in very small packages, I haven't found anything like a SOT23-8, for example. Apart from the TI LM3S101 in a Fred Flintstone Package (aka SOIC-28) most packages seem to be QFP and QFN, and more of the former than the latter.
This is somewhat surprising, since the PCB assembly technology for both is the same, both can be inspected using flying probe, for example (which isn't possible on CSPs). Yet the QFN needs much less space than an equivalent QFP.
The explanation is demand, of course. Apparently most customers don't need the smaller space of the QFN (yet). Some manufacturers are pretty flexible about packaging, and may be prepared to introduce a new package for an existing device if you buy, say, 100k devices per year. This has more administrative than technical implications. So while ARM is widespread most customers will either need smaller quantities or don't really need the new package.
Still I expect ARMs to become available in smaller packages, like less than 20 pins. Especially for Cortex M0 this is going to be needed to successfully take 8-bitters the wind out of the sails. While SOT23 may not be an option I see many possibilities in QFN and particularly DFN.
Unlike DIL DFN is not limited to a specific width. This table
shows how many variants there are available from just 1 manufacturer. So there is always a solution for a specific number of pins and die size.
Small controllers like the LPC1102 would easily fit in a 3 x 3mm QFN-16, for instance, but apparently (and unfortunately?) this hasn't happened yet.
Best Answer
I don't think you'd get any advantage. If you look at a square and an octagon with the same perimeter, the octagon does not look great. On top of the that, you normally lose a bit of space at the corners (compare PLCC and TQFP packages) so I think you'd lose a lot with double the corners.