Read Me
Owing to Haskell’s youth and its origins in academia, research papers are often best for learning some aspects of the language.
See also:

A List of Foundational Haskell Papers by Emily Pillmore.
History
 On Lisp by Paul Graham

Though Lisp is dynamically typed, impure, and eagerly evaluated, this book is worth reading for its historical importance, for its excellent introduction to functional programming and metaprogramming, and for its code excerpts. Translating it to Haskell, or attempting to, is a good exercise.
 Why Functional Programming Matters by John Hughes

This classic presents concrete evidence of the power and simplicity of pure functional programming. Translating the example code to Haskell is another good exercise.
 A History of Haskell: being lazy with class by Paul Hudak, John Hughes, Simon Peyton Jones, and Philip Wadler

The amusing anecdotes within are effective teachers of some of the most technical parts of Haskell. How do data and newtype differ? Why does show [] work in GHCi, but not in a program unless accompanied by a type declaration? Did you know pattern matching in let and where differs to pattern matching in case and lambdas?
 The Glasgow Haskell Compiler by Simon Marlow and Simon Peyton Jones

Fascinating overview of the inner workings of an industrialstrength Haskell compiler. Especially recommended for those who care about performance.
The functional way
I grew up on strictly evaluated impure languages. Some problems seemed difficult to solve without them, until I read more.
 Tacking the Awkward Squad: monadic input/output, concurrency, exceptions, and foreignlanguage calls in Haskell by Simon Peyton Jones

One might think that Haskell is great for research and teaching, but perhaps not as great for practical applications because of messy realworld complications. On the contrary, it turns out Haskell can elegantly resolve these complications.
 Applicative programming with effects by Conor McBride and Ross Paterson

Introduces applicative functors, a useful abstraction lying between functors and monads. Indeed, these are so useful that GHC has elevated applicatives to their rightful place in the pantheon alongside functors and monads.
 Lazy Functional State Threads by John Launchbury and Simon L Peyton Jones

Some data structures work best when mutable, and a clever extension of the type system lets us operate on them in Haskell.
 Purely Functional Data Structures by Chris Okasaki

Although challenging, lazy evaluation and immutability can be helpful when designing data structures. Code is simpler, and we gain persistence. This PhD thesis later became a book.
 Functional Programming with Bananas, Lenses, Envelopes, and Barbed Wire by Erik Meijer, Maarten Fokkinga, and Ross Paterson

They might have gone a bit overboard with the notation, but the underlying idea is incredible. We learn how to manipulate programs algebraically. We calculate programs.
 Beautiful concurrency by Simon Peyton Jones

Introduces Software Transactional Memory (STM), which neatly solves tricky concurrency problems.
 Template Metaprogramming for Haskell by Tim Sheard and Simon Peyton Jones

Sooner or later, we want programs to generate code. This paper shows how it’s done in Haskell.
 Warp by Kazu Yamamoto, Michael Snoyman, and Andreas Voellmy

Haskell’s green threads suit servers, but lazy I/O causes problems. Fortunately, there are clean fixes.
Theory
 Binary Lambda Calculus and Combinatory Logic by John Tromp

Among other things, introduces some theoretical underpinnings of Haskell.
 Proofs are Programs: 19th Century Logic and 21st Century Computing by Philip Wadler

If we care about reliable software, we must study the foundations of mathematics.
 Software Foundations by Benjamin Pierce, Chris Casinghino, Marco Gaboardi, Michael Greenberg, Cătălin Hriţcu, Vilhelm Sjöberg, Brent Yorgey

Functional programming “serves as a bridge between logic and computer science” which can ultimately guarantee bugfree code.
 Theorems for free! by Philip Wadler

Thanks to purity and strong types, we can deduce many facts about a function from its type signature alone.
 Proofs and Types by JeanYves Girard, translated and with appendices by Paul Taylor and Yves Lafont

Just the introduction made me wonder what the equals sign really means.
 Homotopy Type Theory

Perhaps the future of functional programming, and mathematics. According to Robert Harper, if you’re a computer scientist, studying anything else is a waste of time! So now they do know what the equals sign really means?
Compilers
 Lecture notes on how a Haskell compiler works by David Terei, and on the RTS by Edward Z. Yang

Good overviews.
 Implementing Functional Languages by Simon Peyton Jones

How Haskell was once compiled.
 λ to SKI, Semantically by Oleg Kiselyov

An alternative to lambdalifting and supercombinators.
 Typing Haskell in Haskell by Mark P. Jones

How to implement typeclasses.
 Write You a Haskell by Stephen Diehl

A promising book.
 Unboxed values as firstclass citizens in a nonstrict functional language by Simon Peyton Jones and John Launchbury

Use types to help implement unboxed values.
Nonfree

To Mock a Mockingbird by Raymond Smullyan: playful introduction to combinatory logic.

Types and Programming Languages by Benjamin C. Pierce: so widely adored that it is often referred to as "TAPL".

Pearls of Functional Algorithm Design by Richard Bird.

The Algebra of Programming by Richard Bird and Oege de Moor.