If I rewind to a decade ago, not long after I’d first become part of this exciting industry, selection of the most suitable flash memory for an embedded application seemed much simpler. As an off-the-shelf user, selection usually depended predominantly upon the interface you had available, whilst custom developers consulted their mechanical constraints and checked the current cost per megabyte to derive a decision.
Back then, flash technology was fairly uniform – it was just a case of how it was packaged; additionally the types of packaging were fairly small in number. Today, even immediately from memory I could recount 15 different package types, with many of those having multiple variations, such as orientation and the completeness of packaging enclosing the circuitry.
How this plethora of options has come to be is interesting. When flash memory’s high pricing restricted it to industrial applications, there were fewer choices and all were suitable. Once prices fell, flash memory found itself in commercial applications and the technology became more varied and more ambiguous to the embedded developer.
MLC and SLC
The most obvious and popular example of this is the CompactFlash card, born out of the need for compact storage in the emerging market of digital cameras, which was quickly jumped on by the embedded industry. At the time the rule was hard and fast: MLC should remain in the digital camera, and SLC was the only option for embedded operating system storage.
To summarise the difference, Single-Level Cell (SLC) uses a single cell to store one bit of data, whereas Multi-Level Cell (MLC) can process up to four digital states from a single cell. This increased density results in less cells needed per megabyte, thus reduces cost substantially – the flip side is, being written to exponentially more, wearing out is accelerated.
“Strictly SLC” was the de facto policy in embedded computing for a number of years, emphasized so heavily by manufacturers that we were often led to believe MLC was suitable only for the occasional family snapshot – so the first emergence of MLC-based Solid State Drives (SSDs) in laptops some years ago took me a little by surprise.
It quickly became apparent that whilst some manufacturers continued to bang the SLC drum, others had focused on getting more out of MLC and resolving its shortfalls. Today we find that, in fact, the flash controller is significantly more influential in lifetime than the fundamental flash technology.
Current flash controllers are exceptionally intelligent, employing a number of tactics to reduce the volume of data, number of writes, and manage that eventual sector failure to insignificance. Technology such as wear-levelling with bad sector management and internal redundancy alongside write optimization including deduplication and write buffers have bridged that gap whilst retaining the attractive low cost benefit.
In a bid to always have the “perfect product” for an application, at present we’re not seeing the standardization I hoped we would have by now – quite the opposite in fact. Do we really need more than 10 different types of SATA Disk On Module (DOM)?
My concern here is unnecessary variation always results in decreases in stock availability and longevity, with an increase in cost across the industry by side stepping logical economies of scale.
As flash is now such a competitive market, the majority of this glut of package types are actual proprietary formats introduced at OEM level in a bid to secure the supply of the necessary storage medium as it, of course, becomes single source.
Is this a good thing? Probably not – though it would be difficult to convince any OEM responsible that giving away a significant part of their Bill of Materials (BOM) to the lowest bidder for the common good is a sound business strategy.
Flash’s exponential growth
Let’s not forget, the advantages we’ve made in flash memory – even just over a decade – have been exceptional and we do owe some debt to the commercial industry for demanding the quantities necessary to drive that innovation forward. I recall a demonstration of the few Mb “high capacity” DiskOnChips, costing thousands of pounds; then I observed a 64 Gb MicroSD for a fiftieth of the price and fraction of the size. I look forward to seeing what’s next for flash technology.