Every computer, phone, and data center depends on two fundamentally different kinds of memory: DRAM, which holds data only while powered and feeds the processor at high speed, and NAND flash, which retains data without power and stores everything from photos to operating systems. Both are built from enormous grids of simple repeating cells, yet the way each cell works, and the tradeoffs that follow, could hardly be more different. Understanding that difference explains why your laptop has both a modest amount of fast RAM and a much larger, slower solid-state drive, rather than one universal memory technology doing both jobs.
The DRAM Cell: A Capacitor and a Switch
A single bit of DRAM, or dynamic random-access memory, is stored in one of the simplest structures in all of electronics: a tiny capacitor paired with a single access transistor. The capacitor either holds a charge, representing a logical one, or is discharged, representing a logical zero. The transistor acts as a gate, connecting that capacitor to a bit line only when a corresponding word line is activated, allowing a specific row of cells to be read or written while all other rows remain electrically isolated.
This one-transistor-one-capacitor design, commonly abbreviated 1T1C, is prized for its extreme simplicity and compactness, which is why DRAM can pack billions of cells into a small area at relatively low cost per bit. But the design has an inherent weakness: capacitors leak charge over time, even when nothing is actively being done to the cell. Left alone, a charged capacitor representing a one will gradually drain down and eventually be misread as a zero.
Why DRAM Needs Constant Refreshing
To counteract this leakage, DRAM controllers must periodically read the contents of every row and rewrite them at full strength, a process called refresh that typically happens many times per second for every row in the chip. This refresh cycle is the origin of the word “dynamic” in DRAM’s name, distinguishing it from static RAM, which holds its state as long as power is supplied without needing refresh, but which requires far more transistors per bit and is consequently much more expensive to build at large scale.
Refresh overhead also explains some of DRAM’s performance characteristics: a row that is in the middle of being refreshed cannot simultaneously be read or written, and memory controllers must carefully schedule refresh operations to minimize their impact on system performance, a balancing act that becomes more delicate as DRAM density increases and more rows must be refreshed within the same time window.
Reading and Writing a DRAM Array
DRAM cells are organized into a grid of rows and columns. To read a cell, the controller activates its word line, which connects the storage capacitor to a bit line shared by an entire column of cells. Because the capacitor’s charge is minuscule compared to the capacitance of the bit line itself, reading a DRAM cell is actually a destructive operation: the act of sensing the charge disturbs it. A component called a sense amplifier detects the tiny voltage change on the bit line, amplifies it into a clean digital signal, and then immediately rewrites the correct value back into the cell, restoring what reading had just disturbed. This read-sense-restore sequence happens on every single access and is fundamental to how DRAM operates.
The NAND Flash Cell: Trapped Charge as Memory
NAND flash takes an entirely different approach, storing information as electrical charge trapped inside an insulated region of a transistor rather than in a separate capacitor. The classic flash cell design uses a floating gate, a layer of conductive material completely surrounded by insulating oxide, sandwiched between the transistor’s control gate and its channel. Because the floating gate is electrically isolated on all sides, charge placed onto it has nowhere to leak away through normal conduction, which is precisely why flash memory retains data for years without any power applied at all.
Writing to a flash cell, a process called programming, involves applying a strong voltage to the control gate that forces electrons to tunnel through the thin oxide barrier and accumulate on the floating gate, a mechanism known as Fowler-Nordheim tunneling. The presence or absence of that trapped charge shifts the voltage threshold at which the transistor switches on, and reading the cell simply means applying a test voltage and checking whether the transistor conducts, which reveals whether charge is present without disturbing it.
Why Flash Must Be Erased in Blocks
Removing charge from a floating gate, erasing, requires pulling electrons back off through the same tunneling mechanism, and the physical layout of NAND flash makes this practical only at the level of large blocks of cells rather than individual bits. This is the origin of flash’s famous asymmetry: cells can be programmed individually within a page, but an entire block, often spanning hundreds of kilobytes or more, must be erased together before any of its cells can be reprogrammed. This block-erase requirement is why solid-state drives rely on sophisticated controller software to manage writes, track which blocks are worn, and move data around to avoid repeatedly erasing the same physical location, a practice known as wear leveling.
From Single Bits to Multiple Bits per Cell
Early flash memory stored a single bit per cell, known as single-level cell or SLC, distinguishing only between two charge states. Modern flash dramatically increases storage density by distinguishing many more voltage levels within the same physical cell: multi-level cell, triple-level cell, and quad-level cell designs store two, three, and four bits respectively by reading finer gradations of trapped charge. This multiplies storage capacity for the same number of physical cells, but it comes at a real cost, since narrower voltage margins between adjacent states make the cell more sensitive to noise, charge leakage, and wear, generally reducing both the number of times a cell can be rewritten and how long it reliably retains data.
The Rise of 3D NAND
For years, flash density improved by shrinking cells in a single horizontal plane, but as cells grew smaller, the physical effects that make distinguishing charge states difficult grew worse. The industry’s answer was to stop shrinking cells and instead stack them vertically, building NAND flash with dozens or even hundreds of layers of memory cells on top of one another within a single chip. This architecture, called 3D NAND, sidesteps many of the scaling difficulties of planar flash by using cells that are comparatively large and electrically robust individually, while achieving enormous overall density through sheer vertical layer count, an approach that has become the dominant strategy for flash manufacturers.
Why the Two Technologies Coexist
DRAM’s architecture makes it extremely fast, with access times measured in nanoseconds, but volatile and relatively expensive per bit, making it ideal for the working memory that a processor actively computes with. NAND flash’s trapped-charge architecture makes it nonvolatile and inexpensive per bit at high density, but slower to access and limited in how many times a given cell can be rewritten before it wears out, making it well suited for long-term storage rather than active computation. These complementary strengths, rather than any single memory technology winning out, are why virtually every computing device relies on both.
Conclusion
Beneath the marketing labels of RAM and storage lie two genuinely distinct physical mechanisms for remembering a single bit of information: a leaky capacitor that must be constantly refreshed, and an isolated floating gate that holds its charge for years but can only be erased in large blocks. These underlying cell designs, refined over decades, explain nearly everything about how the two memory types behave, from DRAM’s speed and volatility to flash’s endurance limits and block-based erase cycles, and understanding them makes the broader architecture of modern computers considerably easier to make sense of.
Leave a Reply