A cohesive element formulation connects via nonlinear spring elements the relative displacements between the upper and lower surface to a force per unit area. The element is really two dimensional. Instead of strains, the deformation is in terms of the relative displacements between the upper and lower surfaces interpolated to the Gauss points. Unlike strains, the incoming deformations have units of length. The output of the material model is the force per unit area (LS-DYNA manual: traction) at the Gauss points, acting to oppose the displacement.
There are primarily two solid element formulations in LS-DYNA, which can be used with cohesive material models: ELFORM 19/20. ELFORM 20 will transfer moments between the bonded parts, whereas ELFORM 19 will not. The order of the nodes in defining the element is important. The lower surface is defined by nodes 1-2-3-4 and should be connected to one of the bonded parts, while the upper surface, defined by nodes 5-6-7-8, should be connected to the other.
The pentahedral counterparts to the hexahedral ELFORM 19/20 are ELFORM 21/22. The lower and upper surfaces of these elements are nodes 1-2-3 and 5-6-7, respectively.
In addition to the solid element formulations, there is shell ELFORM 29. This bonds the edges of shell elements, with nodes 1-2 being connected to the first edge and nodes 3-4 to the second.
More reading: Review of LS-DYNA Cohesive Elements.pdf
See also: Cohesive material models