I am a body that has learned the oldest trick a structure knows: to carry a roof not by being deep, but by being curved. Ask me to span a hall as a beam and I need metres of depth and tonnes of glue-laminated bulk. Ask me to span it as a shell and I need only a few centimetres of curved lamination, held in the right geometry, and the load runs through my skin instead of fighting my bones. This is the foundation the whole discipline of timber shells stands on, and it is worth understanding before the next parametric solver seduces you into a shape no carpenter can cut.

←TODAY: The BUGA Wood Pavilion (Heilbronn, 2019) proved the segmented timber shell at inhabitable, code-compliant scale — 376 machined cassettes, each geometrically distinct. →3012: The shell that can be un-clicked, grain-sorted and raised again carries its stored carbon forward across buildings, not just across one lifespan. Fulcrum: A surface thin enough to be efficient is also thin enough to be demountable — lightness and circularity are the same property, read twice.

What it is:

A timber shell is a structure that resists load as a membrane — geometry, not depth, does the work. The internal forces travel in-plane, as compression and tension across the surface, and stay there. The moment a thin shell starts to bend, it starts to fail; the entire art is keeping the forces flat inside the skin. Two families dominate, and they take opposite roads to the same efficiency. The elastic lath gridshell is laid out flat as a regular lattice of slender members and then pushed or lowered into a doubly-curved form — the curvature is stored as elastic pre-stress in the wood itself. The segmented plate shell refuses to bend anything: it approximates double curvature from many flat, prefabricated pieces joined edge-to-edge, each facet stiff, the curve faked well enough by faceting. One shell remembers its bending in its cells; the other forgets bending entirely and lives in its joints.

Why it works:

The physics predates the wood by a generation. Félix Candela, Heinz Isler and Eduardo Torroja proved in thin concrete that a surface only centimetres thick could roof a whole hall, because a well-shaped shell carries load as membrane force — tension and compression in the plane — rather than in bending. Bending demands depth, and depth is weight; membrane action deletes both. A shell does not fight gravity by being deep. It fights by being curved.

Timber reaches that same membrane efficiency by two mechanisms. In the gridshell, the form itself is funicular: a hanging net finds the shape in which a given load hangs in pure tension, and inverting that shape gives a surface in pure compression — exactly the method Frei Otto’s Stuttgart team used at Mannheim, form-finding a sagging mesh of links, photographing it, and building the inversion. In the segmented shell, the honesty moves to the edges. Discretising a smooth curve into flat facets does not come free — it displaces the engineering problem from the surface into the seams, where each joint must transfer in-plane shear and normal force while controlling the parasitic bending that segmentation introduces at every kink. This is why the mature work obsesses over joint patterns rather than the visible silhouette. And it is why buckling is the sovereign risk: a thin surface is only as stable as its stiffest weak point, and one under-stiff node can govern the global stability of the whole. The shell is a collective. It stands or falls together.

Origins:

The lattice surface ran first through steel. Vladimir Shukhov raised the first gridshells for the 1896 Nizhny Novgorod exhibition — the Rotunda that proved a lattice could span. Timber inherited the idea because wood is light and willing to bend. Frei Otto and Carlfried Mutschler built the canonical timber one at the 1975 Mannheim Multihalle, engineered with Ove Arup and the early Buro Happold team: a square net of hemlock laths laid flat on the ground and pushed up into a landscape of domes spanning roughly 60 metres without a column. The method travelled — Otto and Shigeru Ban’s paper-tube Japan Pavilion at Expo 2000 in Hannover, the triple-humped oak gridshell at the Weald & Downland Museum (Chichester, 2017, Edward Cullinan Architects with Buro Happold). Then the contemporary chapter, driven by Achim Menges and Jan Knippers at ICD/ITKE in Stuttgart, swapped the bent-lath net for robotically fabricated segmented plates: the 2015–16 Research Pavilion read its joint pattern straight out of sand-dollar morphology, and the 2019 BUGA Wood Pavilion scaled it to 376 finger-jointed hollow cassettes. Between those two lineages sits the Swiss proof that the gridshell is not only for pavilions: the roughly 80-metre free-form timber roof over the Elephant House at Zurich Zoo (2014, Markus Schietsch Architekten with Walt + Galmarini) — a permanent, occupied building, not a summer show.

In practice:

Atelier: At the desk, the timber shell is a lesson in humility before geometry. The temptation, once the parametric tools are open, is to chase the most sculptural double-curve the solver will produce — and then hand an impossible set of non-planar joints to a carpenter. Work the other way: form-find first, then interrogate every step — can each facet be cut flat, is each node reachable by a five-axis machine, is the file that leaves Rhino the file the CNC will actually run? Mannheim was raised by hand from a flat net; BUGA was locked together from 376 machined cassettes; both stand because someone kept the fabrication logic and the structural logic in the same conversation. Your Monday move: before you model anything, add one Boolean to your Grasshopper canvas — a planarity check on every proposed panel (deviation from best-fit plane, flagged red past your CNC’s tolerance) — and refuse to advance a form the fabricator cannot flat-cut.

Hack:

Invert a hanging cable to read the rib heights of a compression arch before you ever open a solver — the funicular is just a parabola you flip in Z. Feed it Mannheim’s numbers and you get the section back in metres:

import numpy as np
span, sag = 60.0, 12.0                 # Mannheim: ~60 m clear, ~1:5 rise
x = np.linspace(-span/2, span/2, 9)
z = sag * (1 - (2*x/span)**2)          # invert the hanging parabola -> arch
print(np.round(z, 2))                  # rib heights along the span, metres

Change sag and watch the whole force path re-tune: a flatter shell buys headroom and pays for it in thrust at the supports; a deeper one calms the thrust and eats the room. That trade-off is the shell’s entire economy, sitting in one variable.

PAZ has drawn this thread before in the concept library — the Catenary panel walks the same hanging-chain logic from Taq Kasra to the Gateway Arch, and the timber-shell concept panels carry the full form-finding Hack in Kangaroo. What the shell adds to that lineage, in wood, is a body that can be taken apart.

Here the century-scale worry arrives, and it is not about span. The buildings my generation regrets are not the ugly ones — they are the landfill skeletons: composites nobody could separate, adhesives that turned a clean timber cassette into hazardous waste, “integral” joints so cleverly bonded that the shell could never be un-clicked. A segmented plate shell is a gift for circularity precisely because every piece is already flat, unique and machine-indexed — the disassembly map is the fabrication map, run backwards. But only if the joints are mechanical, not glued into permanence. The frontier is not more curvature. It is feeding real timber’s grain, defect maps and moisture behaviour back into the form-finding, and keeping every node reversible, so the wood outlives the building and can be sorted and raised again as something else. When the shell can be un-clicked, the membrane that carries load in-plane also carries carbon forward in time.

So before you fall for the curve, run the flip. Feed your span and your sag into four lines, read the rib heights, and only then let the solver loose — with a planarity flag on every panel and a mechanical, reversible joint on every node. Build the shell your grandchildren can take apart.