What is this?

ivar-simple provides immutable, write-once variables (IVars) in the Data.IVar.Simple module.

It also provides two more experimental channel implementations built on top of IVars,

Data.IVar.Simple.IChan: multi-cast channels with write-once semantics.
Data.IVar.Simple.MIChan: channels with write semantics similar to Control.Concurrent.Chan

Comparison to data-ivar

Both data-ivar and ivar-simple provide a write-once variable. That's where the similarities end:

Reading an IVar with data-ivar is an IO operation. In ivar-simple it's a pure function.
data-ivar provides a Reader monoid, monad, etc. for reading from one of several IVars. ivar-simple has no such functionality.
The data-ivar implementation can, in principle, add arbitrary IO actions that are called when an IVar is written. (This detail is not exposed, however)
Technically, ivar-simple tries for efficiency by exploiting the existing locking structures in the RTS; in particular, reading a full IVar for a second time is as cheap as evaluating a record selector. (I have no performance numbers for this.)