More Vapid Engineering
This isn't actually correct. It depends on the design of the bandgap, but sure, the bog standard topologies you see in all the main references are not that great.
For example, LT has a "low noise" voltage ref LTC6655 which quotes around
50nv/sqrthz flatband noise with an IQ=5ma, at 2.5V outputI have designs in production that can do that in around 60ua. Which, on the basis of noise going with the sqrt of current, means something like 5,000 times better performance.
One needs to throw away all the legacy approaches like Brokow. They are all, essentially, Micky Mouse techniques. I have no idea why the likes of Analog Devices and LT haven't moved on with more sophisticated designs.
The general ideas are floating about in the literature, but there are a several important tweaks needed to get them to be really good in practice. Unfortunately, I cant really post them for obvious reasons, but I will say I need 100dB PRS at 200mV dropout, not the usual lying Vout + 1V often quoted on data sheets.
-- Kevin Aylward