Ideal Memory IP For IoT Applications

The notion that IoT SoCs require high-write-endurance embedded NVM is wrong.


The endless possibilities for IoT applications are driving a new wave of imaginative thinking, and for good reason. Recent surveys estimate the number of IoT devices by 2020 will be in the 50 billion units range. The numbers are far too high to ignore and, hence, we have entered a new era of imaginative and creative thinking!

To put this in perspective, IoT applications are growing many times faster than the world population. They will soon be part of everyone’s life, from our connected cars and wearables to smart TVs and homes to medical and infrastructure devices. But despite its potential, the IoT hardware market will be highly fragmented. So far, no company has developed a device that will sell 1 billion units.

While IoT opens new markets to the semiconductor industry and opportunities for differentiation, the design environment is challenging and designers must consider:

• Cost — Many systems have to be less than $10, which means more off-the-shelf hardware and software.
• Power — An operating lifetime must be measured in years using a single battery supply, which means low standby current and migration to smaller geometries.
• Performance — Scalability to accommodate new features, such as voice command, is critical, which means increased CPU power.
• Security — It must work in mission-critical applications, which means added layers of hardware and software complexity.

Because system solution providers want a common, quick and easily re-targetable hardware platform to serve multiple applications, the differentiation is created in software. This is the reason why memory IP requirements are so important. The on-chip memory must be non-volatile, programmable and low-power with instant-on features. It has to have a small silicon footprint with low manufacturing costs and medium-to-low capacity in megabits, not megabytes. Finally, and most critical, the memory IP must be highly secure to protect software IP and prevent hacking.

Hacking is no small problem. A recent HP study found that 70% of IoT devices are vulnerable to hacking attacks, and chip designers creating new IoT products are not security experts. In one recent study, security is the top obstacle to developing connected devices.

Obviously, the right memory IP will go a long way to thwarting an attack. Knowing which memory IP is right for IoT devices may not be that easy because of some long-held misconceptions. For example, embedded ROM provides the lowest power and cost. While it’s true for fixed non-application code, ROM is commercially not practical for changing application firmware.

Another mistaken belief holds that commodity flash provides the lowest programmable “cost per bit.” External flash may provide the lowest cost per bit, but for a typical sub 1-Mbit IoT application, external flash “cost per bit used” is higher than embedded non-volatile memory (eNVM). It also consumes more power for XIP application or if shadowed in on-chip SRAM. It has increased wake-up time due to copy and reduced security, as well.

The notion that IoT SoCs require high-write-endurance eNVM is a fallacy. Low-cost, low-margin IoT end products cannot afford many software updates. High-write endurance is used only during firmware development if emulation capabilities are inadequate. Otherwise, it unnecessarily taxes every shipped product.

The answer for IoT devices is antifuse eNVM, an ideal choice for low-cost, low-power IoT SoCs manufactured in standard CMOS logic processes. It is affordable, scalable, reliable and secure, while giving designers imaginatively thinking about an IoT application one less thing to be concerned about.