Transistor
Also known as: transistors, FET, FinFET, MOSFET, logic gate, semiconductor switch
A transistor is a tiny electronic component that acts as a switch or amplifier for electrical signals.
Every modern computer chip (CPU, GPU, memory, smartphone processor) is built from billions of transistors. When a transistor switches between 'on' (conducting electricity) and 'off' (blocking it), it represents the 1s and 0s of binary computing. The more transistors a chip contains and the smaller each one is, the more computing power the chip can deliver. Large modern processors contain tens of billions of transistors, each smaller than a virus.
Transistor, explained at three levels
At home
A transistor is essentially a microscopic light switch: it can be turned on (letting electricity flow) or off (blocking it). By combining billions of these switches, a computer chip can process information: on means 1, off means 0, and the chip manipulates these 1s and 0s billions of times per second to run your apps, games, and operating system. The reason phones and computers have gotten dramatically faster over the decades is that engineers have found ways to make transistors smaller and smaller, fitting more of them onto a chip the size of a fingernail.
At work
A transistor is an electrically controlled switch: a small voltage on one terminal decides whether current flows between the other two. Chips wire billions of them into logic and memory, so transistor count and size largely set what a processor can do and how much power it burns. For a business, that shows up as performance per watt, which drives laptop battery life, server electricity bills, and cooling costs.
At home
Transistors in modern chips operate as field-effect transistors (FETs), specifically FinFETs and now gate-all-around (GAA) transistors. A FET has three terminals: source, drain, and gate. The gate voltage controls whether current flows from source to drain. In digital logic, the transistor is either fully on or fully off: the analog amplifier behaviour is used in analog circuits (audio, radio frequency). Each logic gate (AND, OR, NOT) is built from a small number of transistors; complex processors chain millions of logic gates into arithmetic units, memory caches, and control circuits.
At work
Transistor count is a rough but useful proxy for chip capability: more transistors generally mean more CPU cores, larger caches, more GPU shaders, or more specialized processing units. But raw transistor count doesn't directly translate to application performance: architecture (how transistors are arranged and what they compute), memory bandwidth, software optimization, and workload characteristics all affect real-world performance. When evaluating chip performance claims, benchmark results for the specific workloads that matter (database queries, ML inference, video encoding) are more meaningful than transistor count alone.
At home
The transistor scaling problem: as transistors shrink below 5nm, quantum mechanical effects become significant: electrons can tunnel through barriers that should block them (leakage current), and manufacturing variability causes individual transistors to behave differently. These effects increase power leakage (static power consumption) and reduce reliability. Chipmakers address this with new transistor architectures (FinFET, GAA/RibbonFET), new materials (high-k dielectrics, metal gates), and design techniques (error correction, redundancy). The physics of transistor scaling is one reason Moore's Law has slowed at the smallest nodes.
At work
3D integration as the successor to planar transistor scaling: as 2D transistor miniaturization slows, chipmakers are stacking chips vertically (3D stacking) using through-silicon vias (TSVs) or hybrid bonding. HBM (High Bandwidth Memory) stacks DRAM dies vertically for much higher memory bandwidth than planar DRAM. Intel's Foveros and TSMC's SoIC stack compute dies with logic dies. AMD's 3D V-Cache technology stacks additional SRAM cache directly on top of CPU compute dies, improving performance in cache-sensitive workloads. 3D integration allows combining chiplets made in different process nodes: a cost-effective alternative to fabricating an entire large chip in the most advanced (expensive) process.
Real-world examples
At home
Sophie was curious why her new laptop felt so much faster than her 5-year-old one, even though they had 'the same amount of RAM.' She learned that the new laptop's chip was built on a 4nm process while the old one used 10nm: meaning the new chip packed far more transistors into the same space. More transistors meant more CPU cores, larger caches, and more efficient circuits that accomplished the same tasks using less power and time. The transistor count difference explained the speed jump without any change in RAM.
At work
Thomas was evaluating two server processor options for a data centre refresh. The newer chip had 3x the transistor count but cost 40% more per unit. He analyzed the relevant benchmarks: the newer chip showed 65% better performance per watt on the database workloads he ran. Over a 5-year server lifecycle, the energy savings (lower electricity and cooling costs) more than offset the premium purchase price. The transistor density difference translated directly into an infrastructure economics decision.
Common confusions
- Many people assume that the vacuum tube and transistor serve completely different purposes — Vacuum tubes and transistors perform the same fundamental functions: amplifying signals and switching between on/off states. Vacuum tubes preceded transistors and enabled early radio, amplifiers, and the first computers (like ENIAC, which used 18,000 vacuum tubes). Transistors replaced tubes because they are smaller by orders of magnitude, require no warm-up time, consume far less power, generate less heat, are mechanically robust, and cost far less to manufacture at scale. The functional equivalence is why tube amplifiers (which audio enthusiasts prefer for their harmonic distortion characteristics) and transistor amplifiers can substitute for each other in audio systems despite the 75-year technology gap.
- Some assume transistor count is the primary determinant of chip performance — Transistor count enables performance but doesn't guarantee it. Architecture matters enormously: the arrangement of transistors into cores, caches, execution units, and interconnects determines what computations the chip can perform efficiently. A chip with 50 billion transistors optimized for mobile power efficiency may perform worse than a 20-billion-transistor chip designed for server throughput on certain workloads. Software optimization, memory architecture, interconnect bandwidth, and the match between chip design and application workload all affect real-world performance more directly than raw transistor count.
For analysts
Understanding transistors at a conceptual level helps in technology vendor discussions, hardware selection, and interpreting performance claims. The practical implications for business analysis: processor generation and process node (not just clock speed or core count) are meaningful differentiators in hardware selection; energy efficiency (performance per watt) is increasingly important for data centre and mobile device economics; and chip architecture differences (explained partly by transistor arrangement) explain why two processors with similar specs perform differently on specific workloads. When clients are making hardware infrastructure decisions, benchmarking on representative workloads matters more than comparing spec-sheet transistor counts.
History & origin
The transistor was invented at Bell Labs in 1947 by John Bardeen, Walter Brattain, and William Shockley: earning them the Nobel Prize in Physics in 1956. It replaced the vacuum tube, which performed the same switching function but was bulky, hot, fragile, and power-hungry. The first transistor was the size of a palm; today's transistors are measured in nanometres (billionths of a metre). Intel's first commercial microprocessor (the 4004, 1971) contained 2,300 transistors. Apple's top M-series Ultra chips contain more than 100 billion. The relentless miniaturization of transistors is the physical manifestation of Moore's Law and the reason computing power has grown so dramatically over 75 years.