A sorting network that first creates bitonic sequences (ascending then descending), then merges them. All comparators at the same depth can run in parallel on hardware.
Worst: O(n log²n)
Average: O(n log²n)
Best: O(n log²n)
Space: O(n log²n)
Parallel: O(log²n)
Ideal for GPU sorting: fixed comparison pattern regardless of data.