Why Your Fast SSD Slows Down After Six Months

Your brand new solid-state drive screams. Half a million random reads per second. Six months later, write speeds crater and latency spikes appear seemingly at random. You check diagnostics. The drive isn't failing. It's doing exactly what the engineering team designed it to do.

This degradation isn't a defect. It's physics meeting economics in the form of silicon and trapped electrons. Tech educator Nikola recently published a detailed breakdown of SSD internals that explains what's actually happening inside that black box on your motherboard. The explanation matters because understanding the mechanism reveals what you can control—and what you can't.

The Architecture of Speed

SSDs replaced spinning platters with three components: a controller (essentially a small CPU running firmware), DRAM cache for mapping tables, and NAND flash chips organized into dies. The parallelism across multiple dies—the controller spreading operations across them simultaneously—delivers the throughput that makes SSDs feel instantaneous compared to hard drives.

But zoom into a single NAND cell and you find a curious design constraint. Each cell is a transistor with a floating gate: a conductive layer sealed inside silicon dioxide insulation. Apply high voltage and electrons tunnel through the oxide onto that gate. Remove the voltage and they're trapped. The insulation won't let them escape even when power is cut. That's your bit storage without electricity.

Reading involves applying lower voltage. Empty floating gate means current flows (binary one). Trapped electrons block current (binary zero). Simple enough. The constraint appears when you try to change that bit.

The Fundamental Mismatch

Cells organize into pages—typically 4 to 16 kilobytes each. Multiple pages form blocks, perhaps 128 or 256 pages per block. Here's the rule that drives everything else: you can read a single page, write a single page if it's already erased, but you can only erase an entire block at once.

Why can't you erase one page? Because every cell in a block sits on shared silicon substrate. The erase voltage hits that substrate and pulls electrons off every floating gate in the block simultaneously. There's no mechanism to erase selectively. Pages must also be written sequentially within a block—out-of-order writing would electrically disturb already-programmed cells.

As Nikola explains: "You can read any page, write pages in order, or erase the whole block. This mismatch between the write unit and the erase unit drives almost every design decision inside an SSD."

Operating systems update individual files constantly. The hardware can't overwrite without destroying adjacent data. Every SSD manufacturer solves this with a Flash Translation Layer.

The Translation Game

Your operating system thinks in logical addresses. Write to file X, update database row Y. It has no idea where data physically lands. The FTL maintains a mapping table—logical address to physical location—stored in that DRAM cache we mentioned.

When you update data, the FTL doesn't overwrite the old page. It writes new data to the next available page and updates the mapping. The old data becomes "stale"—physically present but logically unreachable. This works cleanly at first. But as the drive fills up, you accumulate blocks with mixed valid and stale pages.

Garbage collection reclaims space by copying valid pages to a new location, erasing the entire block, and updating the FTL. Notice what just happened: the OS wrote one update, but the SSD performed multiple writes plus an erase. This ratio—physical writes divided by logical writes—is write amplification.

"On a fresh drive, write amplification is close to one," Nikola notes. "On a full or fragmented drive, it can reach 5, 10, even higher. That's your performance cliff."

The Wear Equation

The oxide insulator degrades each time electrons tunnel through. Consumer drives handle 1,000 to 3,000 erase cycles. Enterprise drives manage several thousand to over 10,000, depending on NAND type. If the same blocks get hammered repeatedly—say, the ones holding database write-ahead logs—those blocks would die first.

Wear leveling solves this by quietly shuffling data. Cold data (rarely changed files) moves onto heavily-used blocks. Hot data moves to fresher blocks. It spreads wear across the entire drive invisibly.

Then there's TRIM. The SSD only sees reads and writes to logical addresses. It doesn't know about your file system. When you delete a file, the OS updates its bookkeeping, but no command reaches the drive. Garbage collection will dutifully copy that dead data to safety, wasting writes and erase cycles on information nobody wants.

TRIM fixes this by notifying the SSD which logical addresses are freed. The drive marks those pages stale in the FTL, just as if they'd been overwritten. Garbage collection skips them. Fewer copies mean lower write amplification and longer drive life.

The Density Trade-off

Single-level cells (SLC) store one bit: empty or full. But storing multiple charge levels per cell increases density. Multi-level cell (MLC) stores two bits with four charge levels. Triple-level cell (TLC) stores three bits with eight levels. Quad-level cell (QLC) packs four bits into 16 levels.

More bits per cell means cheaper storage per gigabit. But charge levels get closer together. Reads slow down, errors increase, and cells wear faster. SLC lasts 50,000 to 100,000 cycles. QLC manages only hundreds to perhaps a thousand.

Most consumer SSDs use TLC—the sweet spot for cost, speed, and endurance. QLC works for read-heavy workloads like cold storage replacing hard drives in data centers. The technology stack reflects economics as much as physics.

What You Control

Every SSD reserves hidden capacity—typically 7 to 28 percent—that the OS never sees. This overprovisioning gives garbage collection room to work. You control the visible capacity. Keep 10 to 20 percent free. The fuller the drive, the harder garbage collection works, the higher write amplification climbs, and the worse your latency becomes.

Steady-state performance matters more than fresh benchmarks. An empty drive is always fast. The question is how it performs after months of fragmentation with garbage collection running constantly in the background.

Monitor drive health through SMART reporting (Self-Monitoring, Analysis and Reporting Technology). Check the percentage-used value. That tells you how much life remains based on actual write patterns, not theoretical maximums.

The performance cliff you experience six months in isn't mysterious once you understand what's happening at the transistor level. Trapped electrons, block-level erasure, translation tables, and wear patterns—these aren't abstractions. They're the physics your data lives inside. The degradation is predictable. More importantly, it's manageable if you work with the hardware instead of against it.

—Bob Reynolds, Senior Technology Correspondent