The “best” (ie high quality) way is to still go with the SDI Decklink option, and put two SDI-to-HDMI converters between the decklink card and your Atem. That way you still get a Fill&Key signal and can overlay your graphic on top of your content.
Then there are varying levels of worse options:
You can go with two UltraStudio Monitors and use them for Fill+Key signals. This has the risk of the Key&Fill signals being unsynced, which will show up as visual artifacts (often moving black edges) when your graphic is moving.
You can also connect your Atem as an external screen to your computers’ HDMI output (remember to configure the screen to a correct resolution and frame rate!) and configure CasparCG to output a “screen consumer” to fill that screen in borderless mode. Then you can do a “green screen” or possibly a “superblack” keying setup to get the transparency. This is of course worse because you’ll get ugly edges (and/or colors) on your graphics - but it’s free, so that’s something…
Here is another thread that touches on this topic: