1.1 — The Physics of Computation

Every photo you've ever taken, every message you've ever sent, every game, every login, every AI conversation — all of it, at the bottom, is one physical trick repeated a few trillion times per second: a tiny switch, flipped not by a finger but by electricity itself.

There are no zeros and ones inside your computer. Open the most powerful chip on Earth and you will find no numbers at all — only silicon, metal, and electric charge sloshing around. The "0" and "1" are a story we agreed to tell about voltages. This chapter is about that agreement: why it was made, why it works, and how a piece of sand was persuaded to do logic.

1. Charge and voltage — the water tower in your pocket

Start with the only physics you need today. (We stay at the level of what electricity does, not why — the atomic story of why materials conduct at all is Volume IV; you need none of it here.)

Matter is made of atoms; atoms carry electrons; electrons carry electric charge — a fundamental property, like mass, measured in coulombs (after Charles-Augustin de Coulomb, who measured how charges attract and repel in the 1780s). Charge is the stuff.

Current is that stuff moving — a flow of charge, measured in amperes (André-Marie Ampère, founder of electrodynamics). One ampere is one coulomb flowing past a point per second. Current is the flow.

Voltage is the pressure pushing that flow — the difference in electric potential energy per unit charge between two points, measured in volts (Alessandro Volta, who built the first battery in 1800). Informally, and perfectly usefully:

The water tower model. Picture a village water tower. The height of the water is voltage — stored pressure, ready to push. The flow through the pipes is current. The narrowness of the pipes is resistance. Nothing flows without height; height without open pipes does nothing.

These three are locked together by the most famous law in electronics, published by Georg Ohm in 1827:

V = I \cdot R

Push harder (more volts) → more flow. Narrower pipe (more ohms) → less flow. This one line explains most of what a wire does, and we lean on it for the rest of the book.

That's the toolkit. Now the question that creates computers: if we want to represent a number with electricity, how should we do it?

2. Why analog fails — the tragedy of the smooth signal

The obvious idea is the analog one: let the voltage be the number. Want to store 0.75? Charge the wire to 0.75 volts. Infinite precision in one wire! Every vinyl record, cassette tape, and radio broadcast of the 20th century worked this way — the signal is an analogy of the thing it represents (Greek analogos, "proportionate").

It has one fatal flaw: the universe won't leave your voltage alone.

Every wire is bathed in interference — neighbouring wires induce stray currents, heat jiggles electrons (thermal noise), power supplies ripple, cosmic rays strike. Your pristine 0.75 V arrives as 0.74. Copy it: 0.72. Amplify and pass it on: 0.76. Every touch adds noise, and in an analog world there is no way to recover the original — 0.74 is a perfectly legitimate value, so nothing can tell it apart from a corrupted 0.75. Errors don't just happen; they accumulate forever. You've heard this: a photocopy of a photocopy of a photocopy, each generation visibly worse.

The digital insight — one of the great ideas of the 20th century — is to stop preserving the value and instead preserve a decision:

Use only two meanings. Voltage near the bottom means 0. Voltage near the top means 1. Everything in between is a no-man's-land that valid signals never rest in.

Figure 1 — The digital bargain. Two wide zones of voltage carry meaning; the gap between belongs to no one. Noise can wiggle a signal anywhere inside its zone without changing what it means. (Voltages illustrative; each chip family sets its own thresholds.)

The width of those safe zones is the noise margin — how much corruption a signal can absorb and still read correctly. It buys the killer feature analog can never have: regeneration. When a slightly-noisy "1" (say 0.83 V) passes through any digital gate, the gate doesn't relay 0.83 — it makes a fresh decision ("that's a 1") and outputs a brand-new, full-strength voltage. Noise is erased at every step. A digital signal can cross the planet through a thousand devices and arrive perfect, because it isn't a value being carried — it's a decision being re-made, again and again.

That is the whole reason your computer is binary. Not because 2 is sacred — but because two levels buy the widest safety zones, and wide zones are what let us build reliable machines from trillions of imperfect parts. (Why not 3 levels, or 10? You'll defend that in this chapter's drill — and meet the one modern place engineers do pack many levels into a single cell.)

One thing is still missing. "A decision re-made at every step" needs a physical device that can make decisions: something that lets one voltage control another. We need a switch — but not one flipped by a finger. A switch flipped by electricity itself.

3. The transistor — a switch flipped by electricity

Engineers built three generations of self-flipping switch. Relays (1930s) used an electromagnet to physically yank a metal contact — audible clicks, milliseconds per flip. Vacuum tubes (1940s) moved the job to electrons flying through a glowing glass bulb — far faster, but hot, power-hungry, and about as reliable as a light bulb.

Then, in December 1947 at Bell Labs, John Bardeen and Walter Brattain — in a group led by William Shockley — made a sliver of germanium amplify a signal with no vacuum, no filament, no moving parts. A solid that switched. Colleague John Pierce christened it the transistor — a blend of transfer and resistor: a resistance transferred under the control of another signal. The three shared the 1956 Nobel Prize, and the century pivoted. The variant that conquered the world — the MOSFET (Metal-Oxide-Semiconductor Field-Effect Transistor, Bell Labs, 1960) — is, by count, the most manufactured object in human history: your phone alone holds more of them than there are people on Earth.

For this chapter, here is the only thing you must know about it — and it is genuinely all you need to understand every computer ever built:

A transistor is a switch with three terminals. Two of them (call them the in and out) are the path current wants to take. The third is the control. Put voltage on the control → the path closes, current flows (switch ON). Remove it → the path opens, current stops (switch OFF). One wire's voltage decides whether another wire conducts.

Figure 2 — The transistor as a controlled switch. A voltage on the control terminal decides whether the in→out path conducts. That single property — a switch flipped by a voltage — is the entire foundation of digital logic.

Read that as a sentence: a voltage controls whether current flows — so the output of one switch can flip the next switch. That's the self-flipping property we needed. Decisions now trigger decisions, with no human in the chain, and because nothing physically moves, a modern transistor flips billions of times per second.

Where the physics lives — and why we're not doing it here

How a voltage on that control terminal opens and closes the path is a genuinely beautiful story — but it's a semiconductor-physics story: what a silicon substrate is, what makes a material p-type or n-type, how doping with a few foreign atoms creates mobile charges, how a PN junction forms a depletion region, and how the control voltage grows a conductive channel. Told properly it needs its own diagrams and a chapter of build-up. Pulling it in here would flood this page with questions we couldn't answer yet and bury the one idea that matters — transistor = voltage-controlled switch.

So it has a dedicated home, fully visual, at engineering-college depth: Volume III (Electronics), Part 2 — [2.1 Semiconductor Physics] (substrate, p/n-type, doping, junctions, depletion regions) → [2.3 the BJT in full] → [2.4 the MOSFET in full], where "the channel appears" gets the complete treatment. Follow that detour whenever your curiosity outweighs your patience. Nothing on this page depends on it.

Transistors come in two complementary flavours, and that pairing is the next idea. One flavour (call it the N-switch) closes when its control is high — "give me a 1 and I conduct." The other (the P-switch) is its mirror: it closes when its control is low. Two switches, opposite triggers. Hold that thought.

4. CMOS — the trick that made a billion transistors affordable

Here's a puzzle. The earliest chips used only N-switches, with a resistor to pull each output up toward the supply. It worked — but whenever an output sat at "0", current flowed continuously from the supply through the resistor to ground. Thousands of gates, each quietly leaking heat while doing nothing. Scale that up and your chip is a toaster.

In 1963 Frank Wanlass at Fairchild patented the fix: use both flavours as a team. Complementary MOS — CMOS, the technology in essentially every chip you own. Its simplest circuit is the inverter (a NOT gate), and it deserves 60 seconds of full attention, because every processor on Earth is mostly this picture, repeated:

Figure 3 — The CMOS inverter. One P-switch between the supply and the output; one N-switch between the output and ground; the input wired to both controls. (Proper schematic symbols arrive in Volume III; here the switches are labeled boxes on purpose.)

Trace it with your finger:

• IN = 1 → the N-switch (on-when-high) closes, the P-switch (on-when-low) opens. OUT connects to ground → OUT = 0.
• IN = 0 → mirror image. P-switch closes, N-switch opens. OUT connects to the supply → OUT = 1.

Output is always the opposite of input — logical NOT. But look at what else is true: in both resting states, exactly one switch is open, so there is no complete path from supply to ground. A CMOS gate holding its value draws almost no current. It spends energy only in the instant of switching — the blink when charge must be poured into or drained out of the next gate's control. That's captured in the formula every chip architect lives by:

P_{\text{dynamic}} = \alpha \cdot C \cdot V^2 \cdot f

Power = (fraction of gates switching, \alpha) × (capacitance charged, C) × (voltage, squared) × (clock frequency, f).

Two of those terms deserve plain words. Capacitance (C) is simply how much electric charge something must hold to reach a given voltage — the "size of the bucket" you have to fill. Every wire and every transistor control terminal behaves like a tiny bucket: to switch it from 0 to 1 you must pour charge in, and to switch it back you must drain it out. Bigger bucket (more capacitance) = more charge to move = more energy per switch, and more time to fill, which is why shrinking transistors — making the buckets smaller — historically made chips both faster and more efficient. Frequency (f) is how many times per second the clock ticks (Chapter 1.2), i.e. how often those buckets get filled and drained each second. So the formula reads, in English: energy per switch, times how many switches happen per second. Park this in long-term memory; it quietly runs the industry. That V^2 is why supply voltages fell from 5 V in the 1980s to about 1 V today — halving voltage alone cuts switching power fourfold. It's why your laptop throttles frequency on battery. And it's why chips didn't melt as they grew from thousands of transistors to over 100 billion: gates that aren't switching are, electrically, almost free.

5. From switches to logic — the gates

An inverter computes NOT. For anything richer, we combine switches. Put two N-switches in series on the pull-down path (with the complementary P-switches above): now the output is dragged to 0 only if the first AND the second control is high — otherwise it floats up to 1. The circuit computes "NOT (A AND B)": the NAND gate, four transistors. A cousin with the pull-downs in parallel computes NOR. These little decision-machines are logic gates — gates because they open or close the path a signal may pass.

The standard way to describe exactly what a gate does is a truth table — a table that simply lists every possible combination of inputs and the output each produces. With two inputs there are only four possible combinations (00, 01, 10, 11), so the table is complete: it leaves nothing about the gate's behaviour unstated. Here are the classic four gates (1 = high voltage, 0 = low):

A	B	NOT A	A AND B	A OR B	A NAND B
0	0	1	0	0	1
0	1	1	0	1	1
1	0	0	0	1	1
1	1	0	1	1	0

Now a fact that sounds like a party trick and actually matters: NAND alone can build everything. Tie a NAND's inputs together → NOT. Follow a NAND with that NOT → AND. Invert both inputs first → OR (by a law you'll meet in a moment). Any function of true and false, however monstrous, can be assembled from NAND gates alone. Engineers call it a universal gate — and it's no accident that manufacturing has made the NAND the cheapest, densest thing we build (flash storage is literally "NAND flash").

So: physics gives a voltage-controlled switch → four switches make a NAND → NANDs make anything. What's missing is the language for designing "anything" — a mathematics of true and false.

6. Boole's algebra — logic becomes mathematics

In 1854 George Boole — a self-taught English mathematician, a shoemaker's son teaching in Cork, Ireland — published An Investigation of the Laws of Thought. His claim: human reasoning itself follows algebraic laws. Let variables take only two values, true and false, and reasoning becomes calculation. His contemporaries filed it under curiosities. It waited ninety years for hardware to exist.

Boolean algebra has three primitive operations — AND (A \cdot B), OR (A + B), NOT (\overline{A}) — and a handful of laws, all provable by checking truth tables:

Law	AND form	OR form
Identity	A \cdot 1 = A	A + 0 = A
Annihilation	A \cdot 0 = 0	A + 1 = 1
Idempotence	A \cdot A = A	A + A = A
Complement	A \cdot \overline{A} = 0	A + \overline{A} = 1
Distributivity	A \cdot (B + C) = A\!\cdot\!B + A\!\cdot\!C	A + B\!\cdot\!C = (A+B)(A+C)

Most feel familiar from school algebra — but notice the strangeness of A + A = A and A + 1 = 1: this is not arithmetic, it's logic wearing arithmetic's clothes. The crown jewels are De Morgan's laws (Augustus De Morgan, Boole's friend):

\overline{A \cdot B} = \overline{A} + \overline{B} \qquad\qquad \overline{A + B} = \overline{A} \cdot \overline{B}

In words: negating an AND turns it into an OR of negations, and vice versa. Two consequences. For hardware, AND-logic and OR-logic become interchangeable under inversion — which is exactly why a whole chip can be built from nothing but NAND gates. For you as a programmer, it's a law you already use weekly: !(a && b) is precisely !a || !b. Every time you untangle a gnarly if by flipping it, you are doing De Morgan — algebra from the age of steam, executing in your editor.

The last bolt was fastened in 1937 by a 21-year-old MIT master's student, Claude Shannon, who'd spent a summer wiring relay circuits. His thesis proved the two worlds are one: any Boolean expression can be realized as a switching circuit, and any switching circuit can be analyzed — and simplified — as Boolean algebra. Circuit design stopped being artisanal knot-tying and became mathematics: write the formula, simplify with Boole's laws, build the smaller circuit, save real money. It is routinely called the most consequential master's thesis ever written — and Shannon returns, spectacularly, in Chapter 1.8.

7. The expert lens

What a chip architect sees here that a beginner doesn't:

The forbidden zone is a treaty, not a wall. Nothing physically stops a voltage resting at 0.5 V — the treaty holds only because circuits are designed to sprint through the zone, never linger. There is one way it genuinely breaks: sampling a signal at the exact instant it's mid-transition, producing a gate stuck between meanings. That failure is called metastability; it can't be perfectly eliminated, only made astronomically rare, and it'll make sense in Chapter 1.2 when we add clocks and memory.

Binary is an economic decision, not a law of nature. The number of levels per wire is a dial. Two levels maximize noise margin; more levels pack more data into the same wire at the cost of tighter margins and heavier error correction. Logic chose 2 and never looked back — yet your SSD's flash cells store 3 or 4 bits as 8 or 16 charge levels in one cell, trading speed and error-correction budget for density. Same physics, different point on one trade-off curve. Spotting "this is a noise-margin trade-off in disguise" is a senior reflex, and it recurs from RAM to Wi-Fi (Volume III, Part 7).

Power is the real limiter. Since ~2005 the question shaping CPUs isn't "how many transistors fit?" but "how many can we afford to switch at once?" — \alpha C V^2 f again. Voltage can't fall much below ~0.7 V (margins collapse, and transistors start leaking even when "off"), so clock frequency stalled near 5 GHz and the industry pivoted to multiplying cores. When you write multi-threaded code in Part 2, you are living downstream of this exact physics.

Next chapter: switches that remember. A gate's output vanishes the instant its inputs change — 1.2 wires gates into loops so they can hold a value, then builds registers, the clock, and the first machine that does one thing after another.

Recall

• Voltage is electric "pressure" (potential difference); Ohm's law V = IR ties it to current and resistance.
• Analog encodes values directly and lets noise accumulate forever; digital encodes decisions in two voltage zones split by a forbidden gap, so every gate regenerates a clean signal.
• A transistor is a voltage-controlled switch: a voltage on its control terminal decides whether current flows — so one switch's output flips the next. (The semiconductor physics of how lives in Vol III Part 2.)
• CMOS pairs complementary switches (P-switch on-when-low, N-switch on-when-high) so resting gates draw ~no current; switching power follows P = \alpha C V^2 f.
• Four switches make a NAND; NAND alone builds every logic function; Boole's algebra (1854) + Shannon's thesis (1937) turned circuit design into mathematics.

Self-test: Why can't analog signals be copied losslessly? State the one property of a transistor that makes logic possible. Why does an idle CMOS gate draw almost no power? Write De Morgan's laws from memory — then again in your programming language.

Quiz Bank

FoundationalWhy do computers use two voltage levels instead of storing values as exact voltages?

Exact (analog) voltages are corrupted by noise, and errors accumulate at every copy or amplification with no way to recover the original. Two widely separated levels create large noise margins: any voltage inside a zone still means the same bit, and every gate outputs a fresh, full-strength level — noise is erased at each step (regeneration). Keywords: noise margin, signal regeneration, discrete vs continuous.

FoundationalIn one sentence, what is a transistor and what makes it useful for computing?

A transistor is a voltage-controlled switch with three terminals: a voltage on its control terminal decides whether current flows between the other two. That lets one signal switch another with no moving parts — so decisions can trigger decisions, billions of times per second. (The internal semiconductor mechanism is covered in Vol III Part 2; for logic, the switch behaviour is all that matters.)

FoundationalWhat did Shannon's 1937 master's thesis establish?

That Boolean algebra and switching circuits are equivalent: any logical formula can be built as a circuit of switches, and any switch circuit can be analyzed and simplified algebraically. It turned circuit design from craft into mathematics and is the founding document of digital design.

AppliedWhy doesn't a chip with 100+ billion transistors melt?

Because of CMOS's complementary structure: in either stable state one switch of each pull-up/pull-down pair is open, so no static current path exists from supply to ground. Energy is spent almost only on switching (P = \alpha C V^2 f), and at any instant most gates aren't switching (\alpha is small). This is precisely why CMOS displaced older NMOS-only logic, whose pull-up resistors leaked current continuously.

AppliedWhy is NAND called a universal gate, and why does industry care?

Tie a NAND's inputs → NOT. NAND + NOT → AND. Invert its inputs → OR (De Morgan). Since {AND, OR, NOT} express any truth table, NAND alone suffices for all of logic. Practically, manufacturing can hyper-optimize one tiny 4-transistor cell and stamp it out by the trillion — the economics of the whole semiconductor industry lean on it.

InterviewState De Morgan's laws and give a concrete place you'd use them in code.

\overline{A \cdot B} = \overline{A} + \overline{B} and \overline{A + B} = \overline{A} \cdot \overline{B} — negation swaps AND↔OR while negating the operands. In code: rewriting !(isValid && isAuthorized) as !isValid || !isAuthorized, simplifying guard clauses, or pushing a negation through a filter predicate. In hardware: converting AND/OR designs into all-NAND (or all-NOR) implementations.

InterviewWalk through how a CMOS inverter produces NOT — and why the forbidden middle zone matters to it.

Input high → the N-switch (on-when-high) closes and the P-switch opens → output tied to ground → 0. Input low → mirror: P-switch closes, N-switch opens → output tied to supply → 1. If the input rested mid-zone, both switches would partly conduct, burning current in a supply-to-ground path and giving an ambiguous output. Valid circuits transit the zone fast; a signal sampled mid-transition causes metastability (Chapter 1.2).

StaffSupply voltages fell from 5 V (1980s) to ~1 V today. Explain the driver and the floor.

Driver: dynamic power scales with V^2, so 5 V → 1 V cuts switching energy ~25×, which made billions of transistors thermally feasible. Floor: below ~0.7 V, noise margins shrink toward the noise floor and transistors leak significantly even when off, so reliability and static power push back. The stall of voltage scaling (~2005) is why clock frequency plateaued and the industry pivoted to multi-core — physics you inherit every time you write concurrent code.

StaffA colleague proposes 4 voltage levels per wire to double density on an on-chip bus. Evaluate.

Frame it as a noise-margin trade. Four levels in the same voltage swing → margins roughly ⅓ as wide → far higher error rates, plus more complex analog-ish drivers and tighter tolerance on every process variation. For logic, regeneration at every gate is the whole value proposition — multi-level breaks it cheaply. But cite the precedent: flash memory (MLC/TLC/QLC) stores 2–4 bits as charge levels per cell, accepted because storage tolerates slow, heavily error-corrected reads. Verdict: wrong for logic paths, proven for dense storage — the answer is the trade-off, not a yes/no.

Flashcards

FlashOhm's law

V = I \cdot R — voltage = current × resistance.

FlashEtymology: “transistor”

Transfer + resistor (John Pierce, Bell Labs, 1947) — a resistance controlled by another signal.

FlashThe one property of a transistor that enables logic

It's a voltage-controlled switch: a voltage on the control terminal opens/closes the current path, so one signal switches another.

FlashThe C in CMOS

Complementary — each gate pairs a P-switch (closes when input is 0) with an N-switch (closes when input is 1); one is always open at rest, so no static current flows.

FlashNoise margin

The width of the voltage zone within which a corrupted signal still reads as the intended bit.

FlashDynamic power formula

P = \alpha C V^2 f — activity × capacitance × voltage² × frequency.

FlashWhy NAND is “universal”

Every logic function can be built from NAND gates alone (NAND → NOT, AND, OR, …).

FlashTwo dates: Boole / Shannon

1854 (An Investigation of the Laws of Thought) / 1937 (Shannon's MIT thesis linking Boolean algebra to switching circuits).

Scenario Drill

DrillA batch of IoT boards works on the bench but corrupts data on the factory floor, near heavy motors. Using only this chapter's physics, form a hypothesis and two mitigations.

Hypothesis: the motors radiate electromagnetic interference that induces noise exceeding the boards' noise margins — signals get pushed across the forbidden zone, so occasional bits flip. The bench is quiet; the floor is not. Mitigations from first principles: (1) widen the margins — run the affected links at higher signal voltage, or slow the edges/clock so sampling happens well clear of transitions; (2) stop trusting single decisions — add redundancy so flipped bits are detected/corrected (checksums or error-correcting codes, formalized in Chapter 1.8). Real teams also shield cables and separate grounds (Volume III territory). The interviewer's checkmark: you tied a field failure to noise margin, not to software.