diagrammatic reasoning for asynchronous circuits
play

Diagrammatic Reasoning for Asynchronous Circuits Dan R. Ghica - PowerPoint PPT Presentation

Diagrammatic Reasoning for Asynchronous Circuits Dan R. Ghica University of Birmingham Samson@60 Oxford, May 2013 How I met your Samson How I met your Samson why async? very high speed (+) very low power (+) high footprint (-)


  1. Diagrammatic Reasoning for Asynchronous Circuits Dan R. Ghica University of Birmingham Samson@60 Oxford, May 2013

  2. How I met your Samson

  3. How I met your Samson

  4. why async? • very high speed (+) • very low power (+) • high footprint (-) • design and verification (-)

  5. The Muller C-element is the typical synchronisation gate. It pro- C duces an output if it receives signals on both inputs. The exclusive or is a merging gate, which outputs if it receives a X signal on either input. The toggle gate alternates (deterministically or nondeterministi- T cally) between the two outputs whenever it receives an input. The forking wire can be seen as a gate which duplicates its input signal.

  6. Ebergen’s model − ⇥ ( AA 0 ) ⇤ ⇤ J W : A → A 0 K = . ⇥ (( A 1 | A 2 ) · A 0 ) ⇤ ⇤ J C : A 1 ⊗ A 2 → A 0 K = ( A 1 A 0 + A 2 A 0 ) ⇤ ⇤ ⇥ J X : A 1 ⊗ A 2 → A 0 K = ⇥ 2 ) ⇤ ⇤ J T : A → A 0 1 ⊗ A 0 ( AA 0 1 + AA 0 2 K = ⇥ 2 )) ⇤ ⇤ J F : A → A 0 1 ⊗ A 0 ( A · ( A 0 1 | A 0 2 K = .

  7. we want some kind of monoidal category

  8. Wire is not identity W : A1 → A2 W’ : A2 → A3 W’oW ∋ A1.A1.A3.A3

  9. Wire is not realistic The reason is that the wires in a circuit are not ideal conductors but have capacitance, which acts like an inertial delay. If the signals are too close, they are “absorbed” by the capacitive inertia. A typical glitchy circuit is the one below: X

  10. can we have a ‘nice’ model?

  11. an affine model J C : A 1 ⊗ A 2 → A 0 K = ( A 1 | A 2 ) · A 0 J X : A 1 ⊗ A 2 → A 0 K = A 1 A 0 + A 2 A 0 J T : A → A 0 1 ⊗ A 0 2 K = AA 0 1 + AA 0 2 J F : A → A 0 1 ⊗ A 0 2 K = A · ( A 0 1 | A 0 2 ) J W : A → A 0 K = AA 0 J U : ∅ → A K = ✏ J E : A → ∅ K = A J P : ∅ → A K = A.

  12. a symmetric monoidal category

  13. algebraic structure 1. ( A, X, U ) is a commutative monoid, with T a retract of X . Associativity ( W ⊗ X ); X = ( X ⊗ W ); X . X X = X X Unit ( U ⊗ W ); X = ( W ⊗ U ); X = W . X X = = Commutativity γ A ; X = X . X = X Retract T ; X = W . X = T

  14. more algebraic structure 2. ( A, C, P ) is a commutative monoid with U an absorbing element. Associativity ( W ⊗ C ); C = ( C ⊗ W ); C . C C = C C Unit ( P ⊗ W ); C = ( W ⊗ P ); C = W P C = = C P Commutativity γ A ; C = C . = C C Absorbing element ( W ⊗ U ); C = ( U ⊗ W ); C = U = = C C

  15. yet more algebraic structure 3. ( A, F, E ) is a co-commutative co-monoid, with C a section of F . Co-associativity F ; ( F ⊗ W ) = F ; ( W ⊗ F ) . = Co-unit F ; ( W ⊗ E ) = F ; ( E ⊗ W ) . = = Co-commutativity F ; γ A = F . = Section F ; C = W = C

  16. even more algebraic structure 1. ( A, X, E, F, U ) is a bialgebra. Theorem 3.07 Distributivity X ; F = ( F ⊗ F ); ( W ⊗ γ A ⊗ W ); ( X ⊗ X ) . X X = X Unit E ; F = E ⊗ E . X = Co-unit X ; U = U ⊗ U . =

  17. further algebraic structure 2. ( A, C, F, X ) is a Laplace pairing (in the sense of Rota, as per [9]). ( X ⊗ W ); C = ( W ⊗ W ⊗ F ); ( W ⊗ γ A ⊗ W ); ( C ⊗ C ); X. C X = X C C

  18. fake algebraic structure g. T ; C = ∅ = F ; X ) T C F X

  19. proofs are very easy calculations (no iteration)

  20. interleaved model with idealized wires f 0 = ∅ , f k = f | f k � 1 , ! f = S i � 0 f i . Note that if f : X → Y then ! f : X → Y . We define C = ! C , X = ! X , T = ! T , F = ! F , W = ! W , U = ! U , E = ! E , Definition 4.11 We say that f : X → Y, g : Y → Z compose safely if and only if !( f ; g ) = ! f ; ! g . Lemma 4.12 All the compositions in Thms. 3.06 and 3.07 are safe in the sense of Def. 4.11. Lemma 4.13 If f : X → Y, f 0 : X 0 → Y 0 then !( f ⊗ g ) = ! f ⊗ ! g . g. T ; C = ∅ = F ; X )

  21. Theorem 4.14 Asynchronous circuits with an interleaved model form a com- pact closed category, called IdAsy where – composition is defined as in A ff Asy ; – identity is W ; – the structural monoidal morphisms (associator, left identity, right identity, symmetry, unit, co-unit) are obtained by applying ! − to the corresponding structural morphisms in A ff Asy ; – objects are self-dual A ⇤ = A ; – the unit ⌘ A : I → A ⇤ 1 ⊗ A 2 and the co-unit ✏ A : A ⇤ 1 ⊗ A 2 → I have the same sets of traces as the identity W : A 1 → A 2 .

  22. Theorem 4.16 The algebraic structure of A ff Asy is preserved by interleaving ( ! � ) in IdAsy : – ( A, X , U ) is a commutative monoid with T a retract of X . – ( A, C , P ) is a commutative monoid with U an absorbing element. – ( A, F , E ) is a co-commutative co-monoid with C a section of F – ( A, X , E , F , U ) is a bialgebra. – ( A, C , F , X ) is a Laplace pairing.

  23. capacitive wires a more realistic model as J K K = !( AA 0 + AA ). A A’ K

  24. 1. K : A ! A is idempotent, i.e. K ; K = K . Lemma 4.18 n n n n n n

  25. towards physical realism: remove the idealized wire component W

  26. recovering the categorical structure Definition 4.21 The Karoubi envelope of category C , sometimes written Split ( C ) , is the category whose objects are pairs of the form ( A, e ) where A is an object of C and e : A → A is an idempotent of C , and whose morphisms are triples of the form ( e, f, e 0 ) : ( A, e ) → ( A, e 0 ) where f : A → A 0 is a morphism of C satisfying f = e ; f ; e 0 .

  27. K K = c C K K K x X = K K = K t T K K K = K f F

  28. n u m b e r o f s i g n a l s i t d o e s n o t m a t t e r t h a n s o m e o f t h e m a r e l o s t . T heorem 4.25 The category of delay-insensitive asynchronous circuits D I A sy is compact closed with 1. dual objects ( A, K ) ⇤ = ( A ⇤ , K ⇤ ) 2. unit ⌘ A : I → A ⇤ ⊗ A defined as ⌘ A = ⌘ A ; ( K ⇤ ⊗ K ) ; 3. co-unit ✏ A : A ⇤ ⊗ A → I defined as ✏ A = ( K ⇤ ⊗ K ) ; ✏ A . T heorem 4.26 The algebraic structure of A � A sy and I dA sy is preserved in D I A sy :

  29. applications

  30. = CALL c x where x CALL is: x c trace-level reasoning?

  31. unit / counit c x of x monoids / comonoids x c x x compact-closed x c structure c c c section of f

  32. conclusion • very preliminary • normal forms? P = • completeness vs e trace-equivalent to p (pulse), when in fact it is equivalent to u ( Our model instead equates • model of feed-back? X = • causality

Recommend


More recommend