Devices memory limitations

[alecandido]

The #618 is a powerful tool, and it would be good enough for all applications, if it were not for memory limitations.

We do not expect performance differences between unrolling and sweepers (since they are both loading the pulses, and if possible the sweeper is the one requiring extra operations), and not even a memory difference in the measurement storage (they are making the same measurement).

Thus, the limitation is coming from the memory required to store the pulses to send.

However, they might depend on many factors, e.g. the type and length of pulses. Currently, the number of measurements is used to gauge the batches, e.g. in Zurich

qibolab/src/qibolab/instruments/zhinst.py

Lines 1225 to 1226 in ef11164

	def split_batches(self, sequences):
	return batch_max_readout(sequences, MAX_MEASUREMENTS)

but this might be just an approximate metric.

So, we should properly understand which are the actual memory limitations, device by device.

[alecandido]

Qblox is pretty explicit about some of them:

1. Instruction Memory

Memory storing the instructions of the Q1ASM program to be executed. Commands are either classical or real-time in nature, defining whether they are executed by the classical core or passed to the buffer to be executed by the real-time core. For QCM modules the maximum number of instructions per program is 16384 and for QRM modules the maximum number of instructions per program is 12288.

2. Data Memory

Data memory is a set of registers that are used to store data during the execution of a Q1ASM program. These registers are referenced as R# from 0 to 63.

https://qblox-qblox-instruments.readthedocs-hosted.com/en/main/cluster/q1_sequence_processor.html#instruction-memory