NOR flash connects its memory cells in parallel, so any byte can be read quickly and code can run directly from the chip. NAND flash connects its cells in series and trades that direct access for far more capacity, a lower cost per bit, and much faster bulk writes and erases. NOR is therefore the usual choice for boot code and firmware, NAND for mass data storage. The right choice starts with one question: is this memory primarily for executing code or for storing data?
NOR vs NAND flash starts with how the cells connect
As non-volatile flash memories, both retain data without power. The key difference is the path between a memory cell and the chip’s wiring. In NOR flash, cells connect in parallel to a bit line. With its own drain and source connections, each cell lets the controller address and read a byte or word directly.
In NAND flash, cells are connected in series as a string: one cell’s drain connects to the next cell’s source. By sharing connections, the layout greatly reduces the number of bit lines and ground wires. At a comparable process node, that layout makes a NAND chip about 60% smaller than a comparable NOR chip. At about 4F², a NAND cell occupies less area than a NOR cell at about 10F², where F is the process feature size.
In circuit terms, the names describe the logic. In NOR flash, the parallel cell arrangement functions like a NOR gate: any single conducting cell pulls the bit line low, so the line stays high only when every cell on it is off. In NAND flash, the series string acts like a NAND gate: the bit line is pulled low only when every transistor in the string conducts. For system designers, parallel connections enable direct random access, while series connections favor density.
Speed depends on the operation, not the flash label
For low-latency random reads, NOR provides byte or word access in about 0.1 µs. By contrast, NAND needs about 25 µs to access the first page, then reads sequentially at up to 30 MB/s. Because NAND is addressed in pages rather than bytes, it cannot offer NOR’s direct byte-level access.
| Operation | NOR flash | NAND flash |
|---|---|---|
| Random read | About 0.1 µs per byte or word | About 25 µs to first page access |
| Sequential read | Direct access, suited to code fetches | Up to 30 MB/s after page access |
| Sustained write | About 0.5 MB/s | About 8 MB/s |
| Erase | About 0.6 s for a 64–128 KB sector | About 2 ms for a block |
In this comparison, NAND writes about 16 times faster and erases about 300 times faster. For NOR, the distinct strength is immediate access to a chosen location. Interface choice also shapes throughput: in QSPI mode, SPI NOR reaches up to 66 MB/s and parallel NOR exceeds 200 MB/s; SPI NAND reaches about 80 MB/s, while raw parallel NAND with ONFI 4.0 reaches 200–400 MB/s.
Capacity and cost make NAND the data-storage choice
NOR flash typically tops out at about 2 Gb per chip. NAND spans a far larger range, from 1 Gb to multiple Tb, and a single 3D NAND chip can offer 1 Tb of memory capacity. That density makes NAND flash the standard for SSDs, USB flash drives, memory cards, and phone storage (eMMC, UFS), as well as Linux root file systems, logs, OTA images, and multimedia.
Behind the cost gap is the cell layout. In a 2021 comparison, 1-Gb SPI NAND cost $0.001 per Mb and 256-Mb NOR $0.0156 per Mb: NAND was about 15 times less expensive per bit.
Erase granularity differs as well. In NOR, erase sectors typically range from 4 KB to 64–128 KB. In NAND, a typical block ranges from 128 KB to 2 MB. For small configuration items, NOR’s finer granularity fits; moving and replacing large page-based data suits NAND flash memories.
NAND flash memories need error management from the start
With raw NAND, capacity comes with system responsibilities. Even SLC NAND requires error-correcting code (ECC). During a write, the controller creates an ECC signature and stores it in the spare area. On reading, it recalculates the signature, compares it with the stored value, and corrects data when mismatches occur. The requirement grows as cells shrink: SLC NAND needs 1 bit of ECC per 528 bytes at 4x-nm nodes, 4 bits at 3x-nm, and 8 bits at 2x-nm. NAND is more prone to bit flips than NOR because of its smaller cell geometry and series architecture; NOR is normally used without ECC.
From the factory, NAND ships with marked bad blocks, and additional blocks can become bad during service. A flash translation layer maps logical addresses to physical locations, while bad-block management skips unusable blocks. For new writes, dynamic wear leveling selects less-used free blocks; static wear leveling moves long-lived data so cycles do not accumulate in a small part of the device. Before a block is erased, garbage collection consolidates its valid pages.
Inside managed NAND such as eMMC and UFS, an integrated controller handles ECC, bad-block management, wear leveling, and garbage collection. With raw NAND, the host processor must provide that functionality. At the system level, the added RAM, software, and validation effort can make a small NOR design less expensive even when its chip price is higher.
Endurance varies sharply by cell type. A single-level cell (SLC) stores 1 bit, a multi-level cell (MLC) 2 bits, a triple-level cell (TLC) 3 bits, and a quad-level cell (QLC) 4 bits; each extra bit lowers cost per bit and endurance. Both SLC NOR and SLC NAND reach about 100,000 program/erase cycles. MLC NAND falls to about 3,000–10,000 cycles, TLC NAND to about 1,000–3,000 cycles, and QLC NAND to about 1,000 cycles.
Execute in place explains why NOR flash is still used
Through execute in place (XIP), a processor runs machine code directly from NOR flash instead of copying it into RAM first. In the processor’s memory-mapped address space, instruction fetches act like direct memory accesses rather than SPI command transactions. As a result, required SRAM falls and boot time shortens significantly.
XIP depends on the host: its SPI or SMIF controller must support memory-mapped operation. Because NAND is page-addressed rather than byte-addressed, it cannot support XIP; code stored there must be shadowed into RAM before execution.
This distinction keeps NOR in use for boot code, BIOS/UEFI firmware, microcontroller firmware, automotive ECUs, industrial controllers, and communications infrastructure. In these uses, code is rarely updated but must be readable immediately after power-up.
Choose the flash architecture by workload, then consider a pair
- Choose NOR when code must execute directly from flash via XIP, random reads matter, or the design needs under 128 Mb without bad blocks or ECC overhead.
- Choose NAND when capacity, low cost per bit, bulk writing, and block erase performance matter more than direct code execution.
- Budget for NAND management. Raw NAND requires ECC, bad-block handling, a NAND-capable driver or file system, and RAM for code shadowing.
- Use both when the workload needs both strengths. A small serial NOR flash boots the system and runs the bootloader through XIP, while a larger NAND chip holds the operating system and user data.
For embedded systems that need more than 256 Mb without a wide parallel interface, SPI NAND provides an effective middle ground. It combines NAND density and cost with an 8-pin SPI interface and spans 512 Mb to 8 Gb. SPI NAND does not support XIP, and bad-block management and wear leveling remain necessary, though on-die ECC reduces host computational requirements.
Power consumption changes with active and standby time
During active read operations, NOR consumes less power than NAND, but its standby current is higher. NAND has the lower standby current, while its active write current is higher. Total energy use depends on the operational duty cycle.
In an IoT sensor that spends extended periods asleep and accesses storage infrequently, NAND’s lower standby current can reduce overall power consumption despite its higher active write current. When a system repeatedly fetches executable code, NOR’s lower active read power is preferable.
Three myths that lead to the wrong flash choice
“NAND is always faster”
NAND achieves higher sustained write and erase throughput, but not lower first-access latency. At about 0.1 µs, NOR’s random byte read is faster than NAND’s initial page access of roughly 25 µs.
“NOR is obsolete”
NOR remains the practical flash architecture whenever an embedded system requires XIP and instant-on execution without shadowing code into RAM.
“NAND is unreliable”
At 100,000 program/erase cycles, SLC NAND matches SLC NOR endurance. Reliability is maintained by implementing required ECC, bad-block management, and wear leveling suited to the workload.
Frequently Asked Questions
Is NAND or NOR flash faster?
For random byte or word reads, NOR is faster at about 0.1 µs versus NAND’s roughly 25 µs page access. For sustained writes and block erases, NAND is faster: about 8 MB/s and 2 ms, compared with NOR’s 0.5 MB/s and 0.6 s.
What are NAND and NOR flash used for?
NOR flash stores code that runs in place: boot loaders, BIOS/UEFI firmware, and microcontroller firmware. NAND flash is used for data storage in SSDs, USB flash drives, memory cards, eMMC and UFS phone storage, file systems, and logs.
Is NOR flash still used?
Yes. For boot code, microcontroller firmware, automotive ECUs, and systems requiring XIP and low-latency random access, NOR remains standard.
Is NOR better than NAND?
Neither is universally superior. NOR is suited for direct code execution and lower capacities under 128 Mb; NAND is suited for high-capacity, low-cost data storage and faster bulk write and erase operations.
Can NAND and NOR flash be used together?
Yes. Many embedded architectures pair a small NOR flash for instant boot via XIP with a larger NAND flash for the operating system and file storage.
How do NAND and NOR flash differ in power consumption?
NOR draws less current during active reads, while NAND draws lower current in standby mode. The more efficient choice depends on sleep versus read/write duty cycles.

