Non-volatile memory (NVM) serves several purposes, including storing processor firmware, encryption keys, PINs, and ID numbers. The two primary NVM technologies in use today are ROM and flash. While both have desirable characteristics, each also has its shortcomings that limit its applicability to many types of new products.
ROM is small, consumes little power, and is implemented in a standard CMOS process. It scales with process-node shrinks, so it may be used as new processes become available. The big disadvantage to using ROM as embedded NVM is that it is mask-programmed and thus configured during the chip’s processing. Any information in an embedded ROM is “locked in” and cannot be changed without modifying the photomask set and re-processing the chip.
The lack of field programmability isn’t always practical for today’s chips, with their accelerated time-to-market requirements. Consumer-product feature sets change very frequently, sometimes as often as every three months. Market demand can then change rapidly for every production version of a device with embedded masked-ROM.
Consider an example where a processor’s firmware is stored in ROM. In a typical system-on-a-chip design flow, firmware is directly in the project’s critical path. It is almost impossible to finish firmware development until the chip hardware is available. This usually means at least one re-spin of one or more of the chip’s masks and the extra time needed to reprocess the chip with the new mask set.
Every change in the contents of a masked ROM, due to firmware modifications, requires a substantial nonrecurring-engineering charge. Furthermore, each change incurs a two- to three-month turnaround or longer in today’s full-capacity fabs. Finally, managing the inventory of chips with masked ROMs, where a different ROM is needed for each variation of the chip (such as for varying feature sets or different firmware), adds substantially to the chip’s cost.
Flash is a mature and proven memory technology in many applications and products. However, embedded flash has several disadvantages for consumer products: high cost, low security, lack of scalability to new processes, and size. Cost-sensitive products such as PDAs, MP3 players, digital cameras and camcorders, games, and cell phones have very low profit margins. Implementing flash on a CMOS logic chip requires additional masks and processing steps, which can add as much as 50% to the chip’s processing cost. Downloading firmware from an external flash chip makes the download process susceptible to undesirable interception by a third party. On-chip flash, through reverse engineering, can be read to reveal the memory’s contents.
One of flash’s biggest drawbacks is in its inability to “keep up” with leading-edge process nodes. By migrating a chip to a smaller process node, the chip vendor can put more functions on the chip, adding to its appeal, or put the current functions on a smaller chip, lowering its cost. Both of these scenarios are very appealing to consumer-product manufacturers. Flash and other charge-storage types of floating-gate NVM technologies are limited in their ability to store charge since, as logic oxides get thinner, direct tunneling occurs and the charge tunnels off.
Consequently, flash charge-storage technologies have scaling limitations between 80 and 85 Angstroms for the wafer’s tunnel oxide thickness, the minimum oxide thickness flash must have to reliably hold a charge once it is stored. Advanced process CMOS gate oxides are currently much thinner, in the range of 30 Angstroms, limiting the implementation of flash to non-leading-edge process nodes.
What we need is a new low-cost, secure, and scalable NVM technology that has high density and low power consumption. The technology should be compatible with a standard-logic process flow and not require additional masks or process steps. High security is a must to protect software IP and prevent theft of stored on-chip information.
This technology exists today and is available for a variety of high-bit-count and low-bit-count applications in the computing, entertainment, and communications market sectors. Using this technology will undoubtedly accelerate the development of new products for applications in these and other markets.
Charles Ng can be reached at [email protected].