Where does stencil collide?

In numerical computing a stencil is the fixed pattern of neighbouring grid points whose values are combined to update a single point, as when a finite-difference scheme approximates a derivative from a cell and its immediate neighbours. The same stencil is swept across the whole array, so performance turns on how its repeated memory accesses map onto cache and parallel hardware. LeVeque's Finite Difference Methods for Ordinary and Partial Differential Equations sets out the underlying mathematics, and Datta and colleagues study the optimisation of stencil computation on multicore machines. The referent is a neighbourhood-update pattern over a discrete grid, sharing only its name with a printmaker's ink mask.[1][2]

Unit

The set of neighbouring grid points combined to update one point

Recovery means

Not applicable

Measured by

Characterised by stencil order and shape and by achieved throughput against available memory bandwidth

In micro- and nanofabrication a stencil is a thin membrane perforated with apertures, held in front of a substrate so that evaporated or sputtered material lands only where the openings allow, patterning the surface in a single step with no resist. Because nothing is coated, exposed, or chemically developed, stencil (or shadow-mask) deposition suits fragile or unconventional materials. Vazquez-Mena and colleagues review this resistless approach and its resolution limits. The referent is a reusable physical shadow mask for additive patterning, distinct from both the printmaker's ink stencil and the computational grid stencil.[3]

Unit

The apertures in a membrane that admit deposited material to a substrate

Recovery means

Not applicable

Measured by

Evaluated by feature resolution, blurring at the aperture edges, and clogging of the mask over repeated depositions