You Can try deleting the impacted controller.
In the past, VMware always recommended deleting the entire control cluster as a precaution when things went bad - even with a single controller. There were some 'quirks' in older builds of NSX that necessitated this. I don't recall the specifics, but I believe the replacement of the entire cluster prevented cluster election issues. In my personal experience, newer builds - 6.3.5 included - should be fine. That said, it's critical that you confirm your control cluster health before deleting a controller (show control-cluster status). Cluster majority is required for the control plane to function in a read/write capacity (i.e. 2 of 3 controllers need to be up). If the other two are completely healthy, you should be able to delete and re-create a single controller node without any noticeable control plane impact. I'd definitely do it during a maintenance window though to be safe.
Hope this helps.My blog: https://vswitchzero.com
Follow me on Twitter: @vswitchzero