folded bit line

Folded bit-line architecture is the DRAM array layout that traded away roughly half the theoretical cell density a memory array could achieve, in exchange for canceling out the single biggest obstacle to reading a DRAM cell at all: the bit line's own stray capacitance dwarfs the tiny storage capacitor it is trying to sense, and the only practical way to see past that imbalance is to make sure every disturbance on the bit line gets canceled before the sense amplifier ever has to decide anything.

Why a Differential Pair Needs a Matched Twin the sense amp reads a difference — noise has to match on both sides Open bit-line Reference pair on a different array Noise on each line is unrelated Folded bit-line True pair runs physically adjacent Coupled noise is nearly identical on both Area cost: every other cell cannot connect to this bit-line pair

The sense amplifier's entire job is to tell which of two nearly-equal voltages is slightly larger, and that comparison only works if everything except the signal being measured matches on both sides. A DRAM cell stores its bit as a tiny amount of charge on a capacitor far smaller than the parasitic capacitance of the long metal bit line it connects to; reading it means dumping that small charge onto the bit line and asking a differential sense amplifier to detect the resulting few-millivolt shift against a reference line held at a known precharge voltage. Any noise that couples differently onto the two input lines — crosstalk from a neighboring line, supply bounce, substrate coupling — looks to the amplifier exactly like a real signal, and a reference line that experiences different noise from the signal line corrupts the comparison before it starts.

Open bit-line architecture, the denser alternative, pairs a bit line in one array with a reference line in a physically separate, unrelated array — and that separation is exactly what breaks the noise cancellation. Because the two lines of the differential pair do not run side by side through the same physical neighborhood, whatever noise each one picks up is uncorrelated rather than shared; the sense amplifier ends up comparing a signal corrupted one way against a reference corrupted a different way, and the difference between them no longer isolates just the stored bit. Open bit-line layouts pack cells more densely, since every cell along a word line can connect to a bit line without sacrificing any to serve as pure reference-side dummies, but that density advantage is precisely what the noise problem makes expensive to exploit at scale.

Density Traded for Signal Integrity folded bit-line pays ~2x area to buy common-mode rejection Open bit-line Maximum cell density per area Poor common-mode noise rejection Folded bit-line ~2x area cost vs. open bit-line Strong common-mode rejection Folded bit-line became standard once shrinking cell capacitance made signal margin, not area, the scarcer resource.

The area cost is real and specific: in a folded bit-line array, only every other cell along a word line can connect to a given bit-line pair, while the alternate cells connect to the neighboring pair instead. This halves the theoretical packing density compared to an open bit-line layout where every cell along the word line connects directly to its own bit line without that alternation. The folded architecture accepts this roughly two-to-one area penalty specifically to keep the true and complementary bit lines of each sensing pair running physically adjacent through the same part of the array, so that whatever noise couples onto one line couples onto its neighbor almost identically — turning a potential differential error into a common-mode disturbance the sense amplifier naturally rejects by construction.

This tradeoff became the practical default not because open bit-line ever got worse, but because storage capacitance kept shrinking while the array's own parasitic bit-line capacitance did not shrink nearly as fast. Signal margin at the sense amplifier is set by the ratio of cell capacitance to bit-line capacitance, and that ratio has been under continuous pressure as DRAM cells scaled down; a design with marginal signal headroom can tolerate far less uncorrelated noise than one with generous headroom. Folded bit-line's common-mode rejection became worth its area penalty precisely as the industry's own scaling trajectory made every millivolt of margin scarcer — the same forces that make every other error-cancellation technique in this glossary series worth its cost once the underlying signal gets small enough to need it.

ArchitectureCell densityNoise rejectionWhy it's chosen
Open bit-lineHigher (every cell connects)Weak — reference pair uncorrelatedHistorically favored when margins were ample
Folded bit-line~Half (alternating connection)Strong — true pair runs adjacentStandard once scaling shrank signal margin

As an actual array layout rather than a table row, the mechanism is visible directly:

Open vs. Folded: The Actual Bit-Line Routing same array, two different physical paths from cell to sense amp Open bit-line array A array B cell BL ref cell BL̄ sense amp BL and BL̄ cross different physical neighborhoods — noise doesn't match Folded bit-line one shared array cell ref cell BL and BL̄ run side by side, folded together sense amp both lines share the same neighborhood — coupled noise cancels in the difference The physical path is the noise-cancellation mechanism — folding costs half the cell density to buy a matched route for free.

Folded bit-line architecture is a direct, physical-layout answer to the same kind of problem this series has covered from several other angles — a signal buried in noise that only cancels if the noise is shared, not just present. Where common-centroid layout cancels a process gradient and autozeroing cancels a time-domain offset, folded bit-line cancels bit-line noise by forcing the differential pair to live through the same physical neighborhood — paying real die area for a correlation the layout would not otherwise guarantee.

Go deeper with CFSGPT

Get AI-powered deep-dives, save terms, and run advanced simulations — free account.

Create Free Account