UK What DRAM is
DRAM (dynamic random-access memory) is a type of memory that stores each bit as an electrical charge in a tiny capacitor, switched by a single transistor. It is the main memory, or RAM, in almost every computer, from an industrial box PC to a data centre server, and it also forms the video memory on graphics cards.
The word dynamic refers to the fact that the charge leaks away within milliseconds. To keep the data, the memory controller reads and rewrites every row on a fixed schedule, called refresh. JEDEC, the body that writes the memory standards, sets the window at 64 ms at normal temperatures, halved when the memory runs hot.
Random access means any location can be read in roughly the same time, unlike a tape or a spinning disk. Robert Dennard at IBM invented the one-transistor DRAM cell in 1966, and the Intel 1103 of 1970 was the first commercial DRAM chip. The basic cell has hardly changed since. What has changed is density: a single modern chip holds billions of cells.
How DRAM works
DRAM chips arrange their cells in a grid of rows and columns, grouped into banks that can work in parallel. To read data, the controller opens a whole row, sense amplifiers detect the tiny charge in each cell and latch it, and the requested columns are sent out over the memory bus. Reading drains the cells, so the row is written back before it closes.
Because the controller has to open rows, wait for them and refresh them, DRAM is many times slower than the SRAM caches inside the processor. It makes up for that with density: one transistor and one capacitor per bit, against six transistors for SRAM, puts gigabytes on a small module at a modest cost per bit.
DRAM is volatile. When power is removed, the charge leaks away and the data is gone within moments, which is why a computer needs separate storage for anything it has to keep.
DRAM vs SRAM
SRAM (static RAM) is the other main kind of random-access memory. The two are built for different jobs, and every computer uses both.
| Feature | DRAM | SRAM |
|---|---|---|
| Memory cell | 1 transistor and 1 capacitor | Usually 6 transistors |
| Refresh | Needed, every row within about 64 ms | Not needed while powered |
| Access time | Typically around 50 to 100 ns from the processor | Around 1 to 10 ns |
| Density and cost per bit | High density, low cost | Low density, high cost |
| Typical capacity | Gigabytes | Kilobytes to tens of megabytes |
| Where it is used | Main memory, graphics memory | Processor caches, registers, microcontrollers |
| Volatile | Yes | Yes |
In short, SRAM is fast and small, so it sits inside the processor as cache. DRAM is large and cheaper, so it provides the gigabytes of working memory behind the cache.
Types of DRAM
All modern DRAM is synchronous (SDRAM), clocked in step with the memory controller. The families differ in how they trade bandwidth, power and capacity.
-
//
DDR SDRAM. The standard main memory, now in its DDR4 and DDR5 generations. Our comparison of DDR4 and DDR5 for industrial computing covers the differences in speed, power and on-die error correction.
-
//
LPDDR. Low-power DRAM soldered next to the processor, used in laptops, tablets and system-on-module designs. NVIDIA Jetson Orin modules use LPDDR5.
-
//
GDDR. Graphics DRAM with very high bandwidth, used as video memory on graphics cards. GDDR6 and GDDR7 are current.
-
//
HBM. High-bandwidth memory, made from DRAM dies stacked on top of each other beside the GPU, used on data centre accelerators.
-
//
ECC DRAM. Standard DRAM with extra bits that detect and correct errors, explained in our quick guide to ECC memory.
DRAM in industrial computers
The memory chips in an industrial computer are the same DRAM as in a desktop, but the way they are fitted and qualified differs.
Many compact boards solder the DRAM directly to the board instead of using a removable module. Soldered memory resists shock and vibration and saves space, at the cost of fixing the capacity at the time of build. Boards with sockets take SO-DIMM modules, which our industrial motherboards use across most form factors.
Temperature matters more than it seems. DRAM leaks charge faster when hot, so above 85 °C it must refresh twice as often, which costs bandwidth. Wide-temperature memory is screened for operation from around −40 to 85 °C, and is the usual choice for fanless and outdoor systems. Error correction adds a further layer for systems that run unattended, either through ECC modules or through in-band ECC, which some Intel embedded processors run on standard memory.
DRAM generations also have a lifecycle. Supply of an older generation tightens as makers move on, which is worth planning for on a long-life deployment.
Quick answers on DRAM
DRAM: common questions
What does DRAM stand for?
Dynamic random-access memory. Dynamic because the stored charge has to be refreshed, and random access because any location can be read directly.
Is DRAM the same as RAM?
Almost always, in practice. RAM is the general term for working memory. In a modern computer that memory is DRAM, so the two words are used for the same modules.
Is DRAM volatile?
Yes. DRAM loses its contents within moments of losing power. Data that must survive a shutdown lives on non-volatile storage such as an SSD.
Why does DRAM need refreshing?
Its capacitors leak. Each cell holds a tiny charge that fades within milliseconds, so the controller rewrites every row on a schedule of about 64 ms.
Is DRAM faster than SRAM?
No. SRAM is faster but far more expensive per bit, which is why it is used for processor caches and DRAM is used for main memory.
We fit the memory in every industrial computer we build, choosing soldered or socketed, wide-temperature and ECC options to match the board and the site, and test it before the system ships.