Photons vs Electrons

Sixty years of computing have been built on exactly one particle. Before this course commits to a second one, it is worth putting their CVs side by side — because every strength and every weakness of photonic computing traces back to four lines in this table.

Property	Electron	Photon
Electric charge	-e	none
Rest mass	9.1\times10^{-31} kg	zero
Quantum statistics	fermion (Pauli exclusion)	boson (happy to bunch)
Mutual interaction	strong (Coulomb)	essentially none in linear media

The previous lesson counted the ways a light wave can carry bits. This one explains, from the particle physics up, why light carries them so well — and why the same physics makes light nearly useless for the other half of computing: switching. Both particles, remember, live double lives as wave and particle; the table above is the particle view of properties whose wave-side consequences we have already met.

Charge: the electron's blessing and curse

The electron's charge is why the transistor exists. A charged particle responds to an electric field, so a small voltage on a gate can raise or lower an energy barrier and choke a current on or off: electrons switching electrons, with gain, a billion times a second, for femtojoules. No other particle offers so convenient a handle.

The same charge is the curse. Moving charge means charging everything nearby — every wire is a capacitor to its neighbours and to ground, roughly 0.2\ \mathrm{pF/mm} on chip, and every bit sent down it costs

E \;=\; \tfrac{1}{2}\,C V^2 \;\approx\; \tfrac{1}{2}\times 0.2\ \mathrm{pF} \times (1\ \mathrm{V})^2 \;=\; 0.1\ \mathrm{pJ\ per\ mm}.

Worse, a wire has resistance as well as capacitance, and an unrepeated wire's delay grows as the square of its length — \tau \sim RC \propto L^2, because both R and C grow with L. Designers fight the square by slicing long wires into segments with repeaters — which restores linear delay but burns energy at every stage. Add crosstalk (your wire's field couples into the neighbours) and frequency-dependent loss, and you have the whole pathology of the interconnect wall in one particle property. The photon has no charge: nothing to charge up, no RC, no crosstalk through the Coulomb field, no resistive loss. A photon launched into a waveguide simply propagates.

Read the chart honestly: over a millimetre or two the repeated wire is perfectly competitive. The photon's line wins on energy and on bandwidth per channel, and its lead in delay only opens up at centimetre scales. This is why photonics attacks the interconnect problem from the outside in — package first, chip later.

Mass and statistics: the perfect courier

The photon's other two CV lines complete the courier's job description. Zero rest mass means a photon always travels at the speed of light in its medium — it cannot slow down, cannot be parked, and needs no acceleration energy; it is pure motion. Bose statistics plus electrical neutrality mean photons ignore each other: millions of WDM channels, and indeed crossing waveguides on a chip, coexist without a whisper of interaction. An electron current cannot cross another wire; a photonic waveguide crossing is a routine, nearly lossless component. Linearity — the wave-level name for "photons don't interact" — is precisely what made the parallelism of the last two lessons possible.

Put a number on the contrast. Sending one bit 10 mm on chip electrically: \sim 1\ \mathrm{pJ} of wire-charging energy. The photons in a detectable optical pulse carrying that same bit: a few femtojoules, and the waveguide adds essentially nothing per millimetre. Run the numbers:

// Electrical wire: C ≈ 0.2 pF/mm at Vdd = 1 V, energy = ½CV² per bit. function wireEnergyPJ(mm: number): number { const cPerMm = 0.2e-12; // farads per mm return 0.5 * cPerMm * mm * 1 * 1 * 1e12; } // Unrepeated RC delay ≈ 0.08·L² ns (L in mm); repeated ≈ 0.15·L ns. const rcDelay = (mm: number): number => 0.08 * mm * mm; const repeated = (mm: number): number => 0.15 * mm; // Photon in a silicon waveguide: group index ~4 → v = c/4 = 7.5e7 m/s. const photonDelay = (mm: number): number => (mm * 1e-3) / 7.5e7 * 1e9; for (const mm of [1, 5, 10, 20]) { console.log( mm + " mm wire: " + wireEnergyPJ(mm).toFixed(2) + " pJ, " + rcDelay(mm).toFixed(2) + " ns unrepeated, " + repeated(mm).toFixed(2) + " ns repeated photon: " + photonDelay(mm).toFixed(3) + " ns"); }

No interaction: the courier who cannot make decisions

Now flip the coin. Computing is not carrying — computing is conditional action: this signal must switch that one. Electrons do it natively; their Coulomb interaction is the transistor. Photons, having no charge, exert essentially no force on one another. In vacuum, photon–photon scattering exists only through exotic quantum corrections, with a cross-section so small that two focused laser beams pass through each other with effectively zero collisions. Light cannot grab light.

The only workaround is to let light act on light through matter: an intense beam slightly alters a material's refractive index, which then affects another beam — the optical Kerr effect and its nonlinear cousins. But these nonlinearities are pitifully weak: where a transistor switches with attojoules of gate charge acting over nanometres, optical nonlinearities demand milliwatts-to-watts of intensity or high-quality resonators to accumulate an effect. This asymmetry — superb carrier, feeble switch — is the single most load-bearing fact in this course. It predicts what history will confirm in the history lesson: attempts to build all-optical transistors (Module 7's post-mortem) have always lost to CMOS, while photonic links and photonic linear algebra — which need no switching in flight — thrive.

Two symmetric misconceptions plague this comparison. The first: "electrons race down wires at nearly light speed." In fact the electrons themselves drift at fractions of a millimetre per second — a leisurely stroll. What travels fast is the electromagnetic field guided by the wire, which rearranges the electron sea at 0.5–0.7c. The signal, in copper as in fibre, is always an electromagnetic wave; the difference is that the copper wave must continually drag charge along with it (hence RC, loss and crosstalk) while the fibre wave travels unencumbered. The second misconception is the mirror image: "photonics wins because photons are faster." As the delay chart shows, they are not meaningfully faster — in a high-index waveguide they can even be slower. The photon's edge has never been velocity. It is that nothing must be charged, dragged, or repeated along the way.

Cross two flashlight beams and nothing happens — but is photon–photon scattering truly zero? Quantum electrodynamics says almost, not quite: two photons can interact via a fleeting virtual electron–positron pair, a process so feeble that for visible light the cross-section is some twenty orders of magnitude below anything measurable in a lab with lamps. It took until 2017 for the ATLAS experiment at the LHC to observe light-by-light scattering directly, using the colossally boosted photon fields around lead nuclei in near-miss collisions. The observation was a triumph for QED — and a wry footnote for optical computing: the universe does permit a photon to switch a photon without any material help, at an interaction strength so absurdly small that you need a particle collider to notice it. Every practical optical logic scheme therefore smuggles matter back in, and inherits matter's speed and energy limits.

Where this goes next

Verdict: the photon is the ideal particle for moving and superposing information, the worst imaginable for deciding. A photonic computer will therefore always be a partnership — photons for transport and linear operations, electrons wherever a decision or a memory is needed. Before drawing the map of that partnership, one more piece of groundwork: what does an optical bit actually cost in energy, end to end? That is the next lesson, and it connects this course directly to Landauer's ledger from reversible computing.