Background
As far as I can tell, the first use of an AVR to do direct digital synthesis of an analog signal was by Jesper Hansen in 2001. Mindaugas Marozas picked up the idea to create a stand-alone function generator in 2006. In 2008, Mindaugas published an improved version, which has become the basis of many homebrew projects, as well as a number of Chinese clones. Georg Latzel, DL6GL, gives a very nice and detailed rundown of the theory of operation (in German).
What’s wrong?
In December 2017, I purchased one of these clones on eBay. Unlike the projects linked above, this one does not require an external three-voltage power supply (+5V for digital logic, -12V/+12V for the op-amp rails). Instead, it uses a single DC barrel connector for (what I thought) +12V, deriving +5V with a 78L05, and -12V with an ICL7660.

Avid readers of the ICL7660 datasheet will already have spotted the problem: The ICL7660 is specified only up to +10V. 1 With a 12V supply, it worked fine for about half an hour, then gave up. Now, the 5V part of the circuit (AVR, display, buttons, high speed digital output) still works fine, but the analog output appears completely dead. On pin 5 of the ICL7660 (VOUT), I measured +0.2ish Volt instead of the expected – V+. At least, the ICL7660 is socketed.2

The Fix
Having replaced the ICL7760, I now get -9V at VOUT, for V+ = +9V, and the analog output works again.
And that’s all?
Fortunately, yes. I also bought a spare opamp, but I’m glad I didn’t need it. I still dread SMD desoldering.
Here are some nice pictures from the working signal generator:
The charge pump at work

Some waveforms





- Using 12V would have required an ICL7760A, which our Chinese friends apparently considered to be too expensive. In small quantities, the price difference is about €0.90 from e.g. Reichelt.
- In fact, the ICL7660 is the only socketed chip on the entire board. Everything else is soldered SMD. Guess someone knew why…