We have used the roleArn and role session name like this:

CREATE STREAMING TABLE table_name
 as SELECT * FROM STREAM read_kinesis (
        streamName => 'stream',
        initialPosition => 'earliest',
        roleArn => 'arn:aws:iam::ACCT_ID:role/ROLE_NAME',
        roleSessionName => 'databricks'
        );

The service principal executing the pipeline has to be able to assume the role referenced by roleArn.