Guiding Light with Total Internal Reflection

Electronics got one enormous favour from nature: electrons are charged, so a copper wire drags them along wherever you bend it. Photons are neutral. Shine a laser across a lab and it goes exactly one way — straight — spreading as it travels. If light is going to compute, the first problem is not logic, arithmetic or memory; it is plumbing. You need a wire for light: something that pins a beam down, steers it around corners a few micrometres wide, and delivers it to the next device with almost nothing lost. The landscape lesson promised chips carrying thousands of optical paths; this module builds that plumbing, and it all rests on one piece of physics you have already met: total internal reflection (TIR). Trap light inside a transparent material that is optically denser than its surroundings, and the boundary becomes a mirror of a kind no polished metal can match — a mirror that is perfect, reflecting 100.000…% of the power, bounce after bounce after bounce.

The perfect mirror, on demand

Recall the rule. When light inside a medium of refractive index n_1 strikes the boundary with a medium of lower index n_2 < n_1, Snell's law n_1 \sin\theta_1 = n_2 \sin\theta_2 tries to bend the transmitted ray away from the normal. Ask for \theta_2 = 90^\circ — the refracted ray skimming along the surface — and you find the threshold angle of incidence:

for light travelling in a medium of index n_1 towards a medium of lower index n_2, the critical angle (measured from the normal) is \theta_c = \arcsin\!\frac{n_2}{n_1};
at any angle of incidence \theta > \theta_c, no power is transmitted: the boundary reflects totally;
below \theta_c the boundary leaks — part of the light refracts out and is lost;
the trick only works going from dense to rare (n_1 > n_2); light arriving from the rare side always gets in.

The word total is doing real work here. A silvered mirror reflects perhaps 95–98% of the light per bounce; after a few hundred bounces almost nothing survives. Total internal reflection is lossless as a matter of principle — energy conservation and Maxwell's equations permit no transmitted wave to carry power away. A signal can bounce a billion times and emerge intact. That single fact is why a glass thread can carry your voice across an ocean, and why this course can contemplate chips where light makes thousands of trips through loops and meshes.

From mirror to wire: the slab waveguide

Now make the mirror into a duct. Take a thin film of high-index material — the core — and sandwich it between lower-index material — the cladding. A ray injected nearly along the film's axis strikes the upper wall at a glancing angle, far beyond \theta_c, and reflects totally; it crosses to the lower wall and reflects totally again; and so on, zig-zagging down the film indefinitely. This sandwich is the dielectric slab waveguide — the simplest wire for light, and the conceptual parent of every optical fibre and every waveguide on every photonic chip. Unlike the hollow metal pipes of microwave engineering, there is no conductor anywhere: the confinement is pure refraction, which is what lets it work at optical frequencies where metals turn lossy.

Notice what the picture implies: only rays within a limited cone of angles around the axis are trapped. The half-angle of that cone, translated to the outside world, defines the guide's numerical aperture, \mathrm{NA} = \sqrt{n_1^2 - n_2^2} — the sine of the steepest external launch angle that still ends up guided. A large index step swallows light from a wide cone; a tiny index step accepts only near-axial rays.

Worked examples: two very different sandwiches

The two material systems that dominate this course sit at opposite ends of the index-contrast scale, and the critical angle tells you immediately how differently they behave.

Telecom fibre — a germanium-doped silica core, n_1 = 1.450, in pure silica cladding, n_2 = 1.444:

\theta_c = \arcsin\frac{1.444}{1.450} = \arcsin(0.9959) \approx 84.8^\circ .

Rays must stay within about 5^\circ of the axis — the guide is fussy, and the numerical aperture is a modest \sqrt{1.450^2 - 1.444^2} \approx 0.13. Bend such a guide too tightly and rays tip below \theta_c and spill out, which is why a bare fibre resents being coiled smaller than a few centimetres.

A silicon photonic waveguide — a silicon core, n_1 = 3.48, clad in silica, n_2 = 1.44:

\theta_c = \arcsin\frac{1.44}{3.48} = \arcsin(0.414) \approx 24.4^\circ .

Now almost any ray is trapped — everything steeper than 24^\circ from the normal reflects totally, a huge 65^\circ-wide cone about the axis. This ferocious confinement is what will later allow bends a few micrometres in radius and components packed by the thousand onto a chip; the price it exacts (fussy coupling, sensitivity to roughness) is a running theme of the lessons ahead. You can check both numbers yourself:

In 1842 the Swiss physicist Daniel Colladon, wanting to illuminate his fluid-jet experiments for a lecture audience, piped sunlight into a tank and let the water escape through a hole. The light, totally internally reflected at the water–air surface, stayed inside the falling jet, bending with it in a glowing arc. John Tyndall repeated the demonstration in London to great acclaim (and rather better publicity), and "light fountains" became a fixture of Victorian spectacle — the illuminated fountains of the great exhibitions were Colladon's trick at garden scale. A jet of water is, of course, a dielectric waveguide: a dense core (water, n \approx 1.33) clad in a rare medium (air), with \theta_c \approx 49^\circ. Every strand of a decorative fibre-optic lamp and every strand under the Atlantic is the same physics — it took 124 years, and a stubborn engineer you will meet two lessons from now, to realise the parlour trick could carry the world's conversations.

The single most common slip in every TIR calculation: the critical angle is measured from the normal — the line perpendicular to the boundary — not from the surface itself. A "glancing" ray that skims almost parallel to the wall has a large angle of incidence (near 90^\circ) and is safely guided; a ray that hits the wall nearly head-on has a small angle of incidence and leaks straight through. If a problem hands you the angle between the ray and the surface, subtract from 90^\circ before comparing with \theta_c. A second, subtler trap: TIR is total in power, but the reflected wave does pick up a phase shift, and the field does not stop dead at the wall — it pokes a short exponential tail into the cladding. Nothing in the ray picture prepares you for that tail; the next lesson takes it seriously, and by the end of the module it will be doing useful work.

Where this goes next

Confinement is the enabling trick of integrated photonics: with light pinned in a core a few hundred nanometres across, you can route it, split it, delay it and interfere it with lithographic precision — everything the coming lessons assemble. But the ray picture you used today is a caricature. Rays at any angle beyond \theta_c seem equally guided, and nothing yet says why a real waveguide supports only a handful of discrete field patterns — or exactly one, if you size it right. To see that, the zig-zag rays must be upgraded to what they really are: interfering waves. That upgrade — modes, and the effective index that prices them — is next.