HF on the ATtiny15

I wanted my class E amplifier to be driven by an Attiny15. The idea is to build a HF (6.78MHz) low voltage and low power beacon.
The maximum PWM frequency attainable using the factory OSCCAL value, and using OCR1B=2 and OCR1A=1 is 8.525MHz with the only possible duty cycle is 33%. With OCR1B=3 we get 6.393MHz, OCR1B=4 we get 5.115MHz.

I want to use my beacons in the ISM band of 6.78MHz so the second value (OCR1B=3) is the closest. I managed to output a duty cycle of 25% (OCR1A=1) or 50% (OCR1A=2), this is perfect for a class E drive as 50% is available. The final tuning was done with trial and error on the OSCCAL value until I got the correct frequency.
In my tests, the factory value for OSCCAL (of all the Atmel Tiny15L I tested) would get you within 1% of the nominal frequency 1.6MHz, but with frequency measurement and tuning I got to within 0.5%.

Did a couple of other idiot measures and observations.
- The program memory of the ATtiny loops, the PC is only 9bits and when the last instruction of Program memory is reached execution continues for 0000.
- A NOP consumes less current that an rjmp +0, not significantly but noticeable. I fill the memory with each instruction and measure the current with or without BOD. The second instruction also take double the time of execution.

RJMP+0,BOD - 3537uA (AVG)
RJMP+0,NO BOD - 3494uA
NOP,BOD - 3503uA
NOP, NO BOD - 3458uA

- OSCCAL after reset is cleared and the measured clock frequency is 1.1MHz, when taken to 0xFF the frequency is 2.083MHz.
- Clock jitter of the oscillator. I filled the program memory with the following:
sbi portb,portb1
cbi portb,portb1
rjmp lp0000

I would then trigger the scope at the low level at the end of the loop. This gap would last longer than normal, then it is easy to trigger on it. You can just see it under the trigger arrow in the picture above.

I would the trigger to the flank increase persistence to max and let measure. the jitter.

The waveform goes up quite fast (in less than 10ns) with the oscilloscope probe as load. With a gate of MOSFET... it's not so good...


Class E - selecting a switch

After the initial results (with a simulated ideal switch) here, had to design a real circuit. Having seen a lot of projects with the 2N7000 and having a few in my surplus bin, I used it in the simulation.
Finding spice models for the 2N7000 is easy, being the only problem that there are so many. I used the typical Coss of 2N7000 (11pf) and subtracted from C1 in the Sokal 2001 formulas.
In order to stop working at the PC for so long, I started using the calculator more often to find optimum parameters and just do a short simulation at the PC. Here's the print out from the calculator.

QL is the Qload of the circuit, K1 is the ratio of XL1 to XC1 (unadjusted) (mentioned in Sokal 2001), one could do some iterative calculations but a simple adjustment seems good enough.
Simulations, unfortunately are not handled (yet) by my HP42S... so here's the ngspice output.
Spice output using 2N7000 Vgs=5V

There was something wrong! After having a closer look at the datasheet of the 2N7000, the drive voltage was clearly a problem. My system is 5V only (Vcc and Vdrive) and it is clearly insufficient to fully drive the 2N7000. Once the drive was changed to 12V the waveforms looked much better.
2N7000 with VGS=12V

Since the system will be 5V only I had to find a logic level drive MOSFET, the ones I had available were IRLD024 or IRLR024N. Unfortunately there are no spice model available for the IRLD024 (Vishay or IR), so I will use the second one (IRLR024N) for simulation and the first one for the final circuit.
Class E with IRLR024N
This last waveform starts showing some signs of out of tune (VDS is 5V at turn-on) but the Sokal documents give hints on how to tune.