Dennard's One-Transistor DRAM Cell: Making Dense Semiconductor Memory Practical

Dynamic random-access memory became the dominant form of computer main memory because it could store a bit in a compact cell built from a transistor and a capacitor. Robert Dennard’s 1967-filed, 1968-issued patent described a field-effect-transistor memory arrangement that helped establish this one-transistor cell. The idea reduced per-bit device count compared with earlier semiconductor memory designs, making higher densities and lower cost per bit feasible as manufacturing improved.

It is tempting to summarize DRAM’s history as “Dennard invented modern memory.” That is too broad. Memory technology developed through competing cell designs, fabrication processes, sense amplifiers, packaging, refresh circuits, address multiplexing, and commercial products. The Computer History Museum’s Silicon Engine timeline places Dennard’s patent in a larger transition: early MOS DRAM challenged magnetic core, and later products adopted the denser cell as technology and economics matured.

Why semiconductor memory competed with core

Before semiconductor RAM became economical, magnetic core memory was a dominant main-memory technology. Core stored bits as magnetic states in small ferrite rings threaded with wires. It was nonvolatile while power was off and could be robust, but assembling tiny cores and threading wires constrained manufacturing scale and cost.

Semiconductor memory promised integration with the transistor manufacturing process. Early MOS memory designs used multiple transistors per bit and still faced tradeoffs among speed, density, and reliability. A smaller cell could put more bits on a die, but a bit had to remain distinguishable under noise, manufacturing variation, and repeated access. The challenge was not only inventing a cell circuit; it was producing enough reliable cells and peripheral circuitry for a useful memory device.

DRAM used a capacitor’s charge to represent data. Capacitors naturally lose charge, so the stored value is temporary and must be refreshed. The “dynamic” label distinguishes this from static RAM, whose bit is held by a transistor latch as long as power is available. DRAM generally traded some latency and refresh complexity for a much smaller cell area and greater density.

The one-transistor, one-capacitor cell

The core DRAM cell contains an access transistor and a storage capacitor. A word line controls the transistor; a bit line connects the cell to peripheral sense circuitry. When the selected word line activates the transistor, charge can move between the cell capacitor and bit line. The sense amplifier detects a small voltage difference and restores the cell’s value. Because reading disturbs or discharges the stored charge, the circuit restores the data as part of the read path and periodically refreshes cells that are not otherwise accessed.

This simple description hides a substantial system. A memory array needs decoders to select rows and columns, sense amplifiers to detect weak signals, timing controls, refresh logic, and interfaces that let a processor read and write values. The cell is the area-defining element, but it cannot operate as useful memory without the surrounding circuitry.

Dennard’s patent, “Field-effect transistor memory,” filed on July 14, 1967 and issued June 4, 1968, describes a transistor-based memory cell. It is a primary record of the design, but a patent filing alone does not tell the whole commercial story. The later move to DRAM dominance depended on fabrication and product engineering by multiple firms.

Why one transistor changed the density equation

Reducing the cell from multiple active devices to one access transistor plus a capacitor lowered the silicon area needed per bit. If a memory chip can fit more cells into a given die area, it can offer greater capacity at a price customers can accept. Smaller cells also increase the importance of manufacturing yield: a larger array has more opportunities for a defective cell, so redundancy and testing become part of the economics.

The one-transistor design also made it possible to scale arrays while sharing peripheral logic. Each bit did not need a full independent amplifier or latch. Rows of cells could use common word lines, and columns could share bit-line circuitry. The resulting architecture depended on high-quality, repeatable processes and careful analog sensing even though the stored information was digital.

DRAM cells faced a physical limit: as a capacitor shrinks, the stored charge becomes harder to distinguish and retain. Later generations used different capacitor structures and process improvements to preserve enough charge per bit. The historical trajectory was therefore not simply “the same cell got smaller forever”; device physics and fabrication required new geometries and materials.

Intel 1103 and the market transition

The Computer History Museum identifies Intel’s 1103 as the first MOS dynamic RAM to present a significant semiconductor challenge to magnetic core memory. Introduced around 1970, the chip contained 1,024 bits. That capacity sounds tiny now, but the important comparison was with core memory’s cost and density at the time. An early commercial chip did not instantly replace core in every computer; it demonstrated a manufacturing path that could keep scaling.

Systems manufacturers had to redesign memory boards, controllers, and power or timing interfaces to adopt semiconductor RAM. The reliability and maintenance profile differed from core, and price per bit changed as yields improved. IBM moved to semiconductor memory in its System/370 Model 158 in 1972, according to CHM’s account, while other companies introduced higher-density parts later.

Mostek’s 4K DRAM of 1973 helped popularize address multiplexing, which reduced the number of package pins by sending row and column addresses over shared input pins at different times. Mostek later adopted the one-transistor cell at 16K density. These developments show why a patentable cell was only one part of an industrial memory transition: packaging, addressing, manufacturing, and compatibility all affected adoption.

Static RAM is not the same tradeoff

SRAM stores a bit using a latch made from several transistors and does not require the periodic refresh cycle used by DRAM. It can be fast and has a simpler read behavior, but each bit consumes more silicon. DRAM’s compact cell made it better suited to large main memories where capacity and cost per bit mattered; SRAM remained useful for caches and other latency-sensitive structures.

The distinction affects how engineers read product specifications. A memory chip’s capacity does not tell whether it is DRAM or SRAM, and neither category alone determines the total system’s speed. A memory hierarchy can use small fast SRAM caches and much larger DRAM main memory. Historical descriptions that say DRAM replaced “computer memory” should specify that it displaced core in many main-memory roles, not that every other memory technology disappeared.

From a research idea to an ecosystem

The one-transistor cell’s impact depended on an ecosystem of semiconductor producers, computer manufacturers, and standards for memory interfaces. Producers worked on cell geometry, process control, packaging, and test. Computer companies adopted parts and modified systems. Memory density increases enabled larger operating systems, more concurrent jobs, and richer applications, which in turn created demand for further density.

Dennard is also associated with the broader idea of scaling transistor dimensions while balancing electric fields, power, and performance. His 1974 scaling paper later became known through a “Dennard scaling” shorthand. That separate concept should not be conflated with the 1968 memory patent: the DRAM cell and transistor-scaling rules are related parts of semiconductor history but answer different engineering questions.

How to interpret the invention claim

Invention claims about memory often compress several milestones: a patent, a laboratory demonstration, a manufacturable component, a commercial product, and adoption at scale. Dennard’s patent supports his key role in the one-transistor cell. CHM’s account credits the cell’s later adoption in denser DRAM generations and identifies products that competed with core. Neither source implies that one inventor alone created the entire modern memory industry.

DRAM’s importance is architectural as much as economic. A smaller bit cell made a large semiconductor main memory practical, and large main memory changed the size of programs and data a system could keep close to its processor. Refresh and sensing introduced costs that designers still manage, but those costs were preferable to the area and manufacturing burden of alternatives for many workloads.

The lasting achievement was a clever division of labor: an extremely small cell stored one bit imperfectly, while shared circuitry repeatedly measured, restored, and coordinated those cells into a reliable array. That bargain between density and maintenance made semiconductor memory scale into the everyday foundation of computing.

Related:

Sources: