John Doyle
Ansys Employee

Below are a few tips to try: 

  1. Drop midside nodes.  The linear shape functions can sometimes offer better stability.  
  2. Consider a coarser mesh.  Not too coarse, but slighly larger elements might be less suceptible to element distortion errors, since their integration points are not so close to the boundary.
  3. Switch back to Aug Lagrange and reduce the contact stiffness.
  4. Use many more substeps.
  5. If the above tips do not help, consider adding Nonlinear Adaptive Meshing.