fundermentals of microelectronics

If you want to design or even genuinely understand analog circuits, this is the standard first stop. Razavi builds every amplifier from semiconductor physics upward, so nothing arrives as a formula without a reason. It suits undergraduates in electrical engineering, self-taught engineers patching gaps, and physicists moving toward chip design. Bring calculus and basic circuit theory; leave with working intuition for gain, impedance, and feedback.

1. 1Chapter 1–2

   Four electrons that run the world

   The book defines microelectronics as the branch of electronics dealing with the design and fabrication of integrated circuits containing millions of components on a single silicon chip. Every phone and laptop depends on getting that right. Then Razavi drops to the atom itself. A silicon atom carries four valence electrons and shares each one with a neighbor, locking the crystal into a lattice of covalent bonds. That perfect sharing leaves the material nearly insulating — until you learn the bandgap is 1.12 eV at room temperature, the exact number that decides whether electrons can break free and carry current.

   The idea

   The whole industry rests on a material that is boring in its pure state. Silicon's four valence electrons create a stable covalent lattice, and the 1.12 eV bandgap sits in a sweet spot: small enough that thermal energy and doping can free carriers, large enough that the crystal doesn't leak current everywhere. That balance is why silicon beat other semiconductors to dominance, and why a chip can hold millions of components without the parts interfering through the substrate.

   Try this: Sketch a 2D grid of silicon atoms sharing valence electrons, then mark where a freed electron and its hole would appear.

2. 2Chapter 2

   Poison the crystal, on purpose

   Chapter 2 asks the question any honest reader would: if pure silicon barely conducts, how does a chip switch billions of times a second? The answer feels like sabotage. Doping silicon with phosphorus (a donor) creates n-type material whose majority-carrier electron concentration is approximately equal to the donor concentration. The crystal's conductivity is now a dial the engineer turns during fabrication.

   The idea

   Doping is the trick that makes semiconductors programmable at the atomic scale. A phosphorus donor contributes one extra electron per atom, so carrier concentration tracks the recipe rather than temperature alone. What surprises newcomers is the precision: a part-per-million change in doping shifts conductivity by orders of magnitude. Every transistor on a chip is, at bottom, a region where somebody chose a number like 10^16 cm^-3 and got exactly the behavior they paid for.

   Try this: Calculate how many dopant atoms sit in a cubic micrometer at 10^16 cm^-3 doping — the count is smaller than you'd guess.

3. 3Chapter 2

   The junction that charges itself

   Press an n-type region against a p-type region and something eerie happens with no battery attached. Electrons diffuse across, ions are exposed, and a depletion region forms carrying an electric field that pushes back. The built-in potential of a silicon pn junction is derived as V0 = (kT/q) ln(N_A N_D / n_i^2), with n_i for silicon given as 1.08 × 10^10 cm^-3 at 300 K. The resulting built-in potential is approximately 0.7–0.8 V depending on doping levels. The diode, the transistor, everything downstream — all of it begins with this self-made barrier.

   The idea

   A pn junction generates its own potential difference from nothing but doping asymmetry and thermal statistics. The logarithm in V0 is the surprise: even enormous changes in doping move the barrier only modestly, which is why silicon junctions cluster near 0.7–0.8 V no matter how you design them. That built-in field is what keeps majority carriers in place, lets reverse bias widen the depletion region, and ultimately gives diodes and transistors their one-way, voltage-controlled character.

   Try this: Compute V0 for N_A = N_D = 10^16 cm^-3 and confirm it lands in the 0.7–0.8 V range.

4. 4Chapter 3

   Diodes by shortcut: the 0.8 V rule

   Chapter 3 faces a practical crisis: the exponential diode equation makes hand analysis miserable. The book presents the constant-voltage diode model, assuming a forward-bias voltage of 0.8 V for silicon diodes in typical circuit analysis. It is immediately put to work on a half-wave rectifier circuit with a 10-V peak input, a 1-kΩ load, and a diode forward drop of 0.8 V, yielding a peak output voltage of 9.2 V. One subtraction replaces a transcendental equation, and the rectifier — the front door of every power supply — falls out of the analysis in two lines.

   The idea

   Engineering modeling means choosing the crudest model that still answers your question. The constant-voltage diode model throws away the smooth exponential and keeps only the cliff edge at 0.8 V, which is enough for rectifier and clipper circuits where the diode is clearly on or off. The 9.2 V peak output shows the payoff: accuracy within a few percent, zero iteration. Knowing when a crude model suffices — and when it collapses — is a skill the book teaches by example rather than sermon.

   Try this: Redo the half-wave rectifier with the exponential model in a spreadsheet and compare the peak against 9.2 V.

5. 5Chapter 4

   The exponential transistor

   Chapter 4 introduces the bipolar transistor, and its governing law is startlingly compact: the collector current is I_C = I_S exp(V_BE / V_T), where V_T = kT/q ≈ 26 mV at room temperature. A few tens of millivolts on the base-emitter junction multiplies the collector current by e — that's the leverage. Then Razavi confronts a flaw. The Early effect: increasing the collector-base reverse bias narrows the base and increases collector current, modeled by the Early voltage V_A, typically 50–100 V, giving I_C = I_S exp(V_BE / V_T)(1 + V_CE / V_A). The transistor obeys an exponential, but not a perfectly clean one.

   The idea

   The BJT is a voltage-controlled current source whose control law is the same exponential as the diode, scaled by 26 mV of thermal voltage. That steepness is a gift — enormous gain from tiny voltage swings — and a liability, since it makes biasing touchy. The Early effect adds a wrinkle: the transistor isn't a perfect current source, because its own output voltage feeds back into its current through base-width modulation. Every finite gain you'll ever measure in a BJT amplifier traces back to that (1 + V_CE / V_A) factor.

   Try this: Using V_T = 26 mV, calculate how much V_BE increase doubles the collector current — about 18 mV.

6. 6Chapter 5

   The workhorse stage that started it all

   Chapter 5 turns the transistor into an amplifier. The common-emitter stage is the first candidate: the book analyzes a common-emitter stage with a collector resistor of 1 kΩ, a supply voltage of 2.5 V, and a bias collector current of 0.5 mA, achieving a small-signal gain of -20 — twenty times bigger, inverted, from a circuit of one transistor and one resistor. Razavi then flips the transistor around. The book derives the input impedance of a common-collector stage as R_in = r_π + (β + 1)R_E, where β is the current gain, typically 100, so a load resistor gets multiplied a hundredfold. One transistor, two personalities: a voltage amplifier and a buffer that barely loads what feeds it.

   The idea

   Gain and impedance are separable design choices, and the common-emitter plus common-collector pair shows why. The emitter stage converts collector current bias into a −20 voltage gain through the 1-kΩ resistor; the follower stage uses the (β + 1)R_E reflection to present an input impedance a hundred times the load. Chaining them lets you amplify without disturbing the source. This division of labor — one stage for gain, one for isolation — is the template for nearly every multi-stage amplifier that follows.

   Try this: With β = 100, compute the input impedance of an emitter follower driving a 1-kΩ load, including r_π.

7. 7Chapter 6

   A transistor made of fields, not junctions

   Chapter 6 swaps the bipolar transistor for the MOSFET, the device that lets chips scale to millions of components. Its behavior splits cleanly in two. In the triode region (V_DS < V_GS − V_TH), I_D = μ_n C_ox (W/L) [(V_GS − V_TH)V_DS − V_DS²/2], acting like a voltage-controlled resistor. Push the drain voltage past that line and it saturates: in saturation, I_D = (1/2) μ_n C_ox (W/L)(V_GS − V_TH)², a clean square-law current source. Then comes a complication the schematic never shows: the body effect — the threshold voltage increases when the source-body junction is reverse-biased, characterized by the body-effect coefficient γ.

   The idea

   The MOSFET's square law is what made digital logic and dense analog integration possible — a gate draws no steady-state current, yet controls the drain through geometry (W/L) and oxide capacitance. The triode split gives designers a resistor and a current source in one package. The body effect is the sobering footnote: the substrate is a fourth terminal you can't ignore, and γ means your carefully chosen threshold drifts whenever source and body potentials diverge, especially in stacked circuits like cascodes.

   Try this: For a device with μ_n C_ox = 100 μA/V², W/L = 10, and overdrive 0.2 V, compute the saturation current.

8. 8Chapter 7

   One transistor, minus ten

   Chapter 7 builds the first real CMOS amplifier. A common-source amplifier with a drain resistor produces a voltage gain of −g_m R_D; e.g., with a 5-kΩ drain resistor, a 1.8-V supply, and a bias current of 0.2 mA, the gain is −10. Simple, effective, and limited — the gain depends on the resistor, and the bias eats headroom. The common-gate stage has a voltage gain A_v = g_m R_D, where g_m = 2I_D / (V_GS − V_TH); for I_D = 0.5 mA and V_GS − V_TH = 0.2 V, g_m = 5 mS. Two stages, two ways to turn bias current into gain.

   The idea

   Transconductance is the currency that converts bias current into voltage gain, and both stages spend it differently. The common-source inverts and offers high input impedance at the gate; the common-gate preserves signal phase and sidesteps the Miller multiplication of capacitance, at the cost of a low input impedance at the source. Notice the book's arithmetic habit: g_m = 2I_D / (V_GS − V_TH) means you can read a stage's gain straight off its bias sheet — 0.5 mA and 0.2 V of overdrive is 5 mS, before any small-signal modeling.

   Try this: Take the 5-mS common-gate device, attach a 5-kΩ drain resistor, and predict the gain before checking the formula.

9. 9Chapter 8

   The op amp as a black box

   Chapter 8 pauses the transistor-level grind and hands you an op amp as a black box: the ideal op amp has infinite input impedance, zero output impedance, and infinite voltage gain, and is typically used with negative feedback. Skeptics will call this cheating — you haven't built the thing yet. Then Razavi runs the numbers. A noninverting amplifier with R1 = 1 kΩ and R2 = 9 kΩ gives a closed-loop gain of 1 + R2/R1 = 10. The infinite open-loop gain, harnessed by feedback, makes the amplifier's accuracy independent of the messy silicon inside. You can design systems before you can design devices.

   The idea

   The ideal op amp is a deliberate fiction that works because negative feedback divides away the imperfections. Infinite gain sounds absurd, but it forces the two inputs to equalize, and once that happens the external resistors alone dictate the closed-loop gain — 10, from 1 kΩ and 9 kΩ, regardless of what the silicon is doing. This is the first place in the book where abstraction beats detail, and it's a lesson in layering: black-box thinking at the system level, transistor thinking underneath when the fiction breaks.

   Try this: Design a noninverting amplifier with a gain of exactly 5 using standard resistor values, then verify with 1 + R2/R1.

10. 10Chapter 9

    Stacking transistors for brute-force impedance

    Chapter 9 confronts a ceiling: a single transistor's output resistance caps the gain you can reach. The cascode is the fix — a cascode stage boosts output impedance by approximately g_m r_o times, where r_o is the transistor output resistance (e.g., 20 kΩ for a BJT at I_C = 0.5 mA), turning kilohms into megohms. Razavi then shows the same stacking philosophy applied to current: a simple MOS current mirror copies a reference current: I_out = I_ref (W/L)_2 / (W/L)_1, assuming equal V_GS, with a typical matching error of 1%. Two transistors sharing a gate become a photocopier for current.

    The idea

    The cascode and the mirror are the same instinct in two directions: use a second transistor to make the first behave better. Cascoding multiplies output impedance by g_m r_o, which is what lets high-gain stages survive without enormous supply voltages. Mirroring exploits matched V_GS to make current copying a ratio of geometry — set the width ratio and you set the copy, accurate to roughly 1%. Both techniques trade headroom and complexity for precision, the central bargain of analog design.

    Try this: Sketch a cascode and label where the g_m r_o multiplication happens; then size a 2:1 mirror for a 100-μA copy.

11. 11Chapter 10

    Two inputs, one stubborn refusal

    Chapter 10 builds the differential pair, the input stage of every op amp. Two transistors share one tail current, and the differential gain of a bipolar pair is A_d = g_m R_C, with g_m = I_EE / (2 V_T); for I_EE = 1 mA, g_m = 19.2 mS per transistor — the pair amplifies the difference between its inputs while ignoring what they share. Except it doesn't ignore it perfectly. The common-mode rejection ratio (CMRR) of a differential pair with a non-ideal tail current source of output resistance R_SS is approximately g_m R_SS. The circuit's whole virtue is a ratio, and the ratio is only as good as the tail.

    The idea

    The differential pair is elegant because symmetry does the work: identical halves cancel common-mode drift, while the difference gets amplified by g_m R_C. The 19.2-mS figure at 1 mA of tail current shows how directly bias translates into performance. But CMRR ≈ g_m R_SS exposes the hidden dependency — rejection isn't a property of the pair alone, it's purchased from the tail current source's quality. This is why current-source design, not transistor design, often decides how good an op amp's input stage really is.

    Try this: Compute the differential gain of a pair with g_m = 19.2 mS and R_C = 5 kΩ, then its CMRR for R_SS = 100 kΩ.

12. 12Chapter 11

    When your own gain sabotages your bandwidth

    Chapter 11 delivers the bad news: amplifiers don't stay amplifiers at high frequency. The villain is Miller's theorem: a feedback capacitor C_F between the input and output of an inverting amplifier with gain −A is equivalent to an input capacitance of (1 + A)C_F. Your small parasitic becomes a monster multiplied by the very gain you worked for. The −3 dB bandwidth of a common-source amplifier with a 10-kΩ load and a total output capacitance of 100 fF is 1 / (2π R_L C_L) ≈ 159 MHz. Gain and bandwidth are in a fight, and the capacitor picks the winner.

    The idea

    Miller multiplication explains why high-gain inverting stages are slow: the (1 + A)C_F effect inflates the input capacitance until the driving source can't charge it fast enough. The 159 MHz figure from a mere 100 fF shows how little capacitance it takes to matter when resistance is in the kilohms. The design escape routes — cascoding to reduce the effective gain per stage, or common-gate topologies that dodge Miller altogether — all trace back to this one theorem. Frequency response is where circuit topology stops being about gain and starts being about capacitance.

    Try this: For a 10-kΩ load, calculate the load capacitance that limits bandwidth to 1 MHz — then find it on a real datasheet.

13. 13Chapter 12

    Feedback: trading gain for control

    Chapter 12 formalizes the trick Chapter 8 only gestured at. A voltage-voltage feedback amplifier with an open-loop gain of 1000 and a feedback factor of 0.1 yields a closed-loop gain of 1000 / (1 + 1000 × 0.1) ≈ 9.9. You gave up almost all your gain — and got something better. Negative feedback reduces gain sensitivity by the factor 1 + loop gain; e.g., a 10% open-loop gain variation becomes a 0.1% closed-loop variation for a loop gain of 100. The transistors can misbehave; the resistors around them hold the line.

    The idea

    Feedback is the most consequential idea in the book because it converts an unreliable amplifier into a reliable system. The closed-loop gain of 9.9 from an open-loop 1000 shows the exchange rate: you spend raw gain to buy precision. The sensitivity result is the deeper point — with loop gain of 100, a 10% drift in the active circuitry becomes 0.1% at the output, a hundredfold improvement. Every stable, manufacturable analog system ever shipped relies on this trade, which is why Razavi spends a full chapter on it before letting you near an op amp's internals.

    Try this: Compute the closed-loop gain if the open-loop gain of 1000 drops to 500 with feedback factor 0.1 — the answer should barely move.

14. 14Chapter 13

    Push, pull, and the 78.5% ceiling

    Chapter 13 faces the end of the signal chain: the output stage, where the circuit must deliver power, not just voltage. A push-pull configuration — one transistor sources the load, the other sinks it — is the natural answer, but left cold it distorts. A class AB output stage with a quiescent current of 1 mA in a push-pull configuration achieves a maximum efficiency of 78.5% under ideal sinusoidal conditions. Not the theoretical maximum of a switching stage, but far beyond a class A design that burns full current around the clock.

    The idea

    Class AB is a negotiated settlement between efficiency and distortion. Pure class A wastes power conducting all the time; pure class B distorts at the zero crossing when one transistor hands off to the other. The 1-mA quiescent bias keeps both devices barely alive, eliminating the crossover glitch while preserving the push-pull efficiency. The 78.5% ceiling under ideal sinusoidal drive is the number that explains why nearly every power output stage — audio amplifiers, drivers, regulators — lives in class AB rather than at either extreme.

    Try this: Compare the 78.5% class AB ceiling against the theoretical class A maximum for the same sinusoidal drive.

Undergraduate EE students taking their first analog electronics course, and self-taught engineers who can compute gain but can't explain where it comes from. Physicists and makers moving toward chip or board-level analog design will find the physics-to-circuit bridge they've been missing.

FAQ