Paul van Gerven
6 September

Thanks to British research, flexible processors have progressed to the point that they can perform useful calculations.

Researchers from Arm and PragmatIC Semiconductor have fabricated the first flexible Arm-based microcontroller, consisting of metal-oxide thin-film transistors on a flexible substrate. Although the performance of the Plasticarm system-on-chip is more or less on par with the most advanced silicon from the early 1980s, that’s still twelve times more complex than the state of the art in flexible electronics.

The two British firms recently published details of the SoC in Nature. The flexible processor was manufactured in a 0.8 μm process, laying down about 18,000 logic gates in an area of 59 mm2. It’s based on the Arm Cortex-M0 architecture and can run ArmV6-M code from its internal memory, which consists of 128 bytes of RAM and 456 bytes of ROM.

Though far from a commercially viable product, the Plasticarm represents a significant step forward for adding ‘intelligence’ to everyday objects, says John Biggs, co-founder of Arm and distinguished engineer at Arm Research. “As ultra-low-cost microprocessors become commercially viable, all sorts of markets will open with interesting use cases such as smart sensors, smart labels and intelligent packaging.”

One example would be a smart label that calculates expiration dates based on how a product (food or prescription drugs) has been stored and handled. Another application could be mesh networks of vibrational sensors installed in water pipes to track down leaks. “Products using these devices could help with sustainability by reducing food waste and promote the circular economy with smart lifecycle tracking,” Biggs points out.

Flexible Arm
Credit: Arm

Thinning down

The flexible electronics toolkit already contained quite a number of components such as sensors, batteries and even displays, but a practical flexible microprocessor was still missing. This is hardly surprising, given the fact that it’s a complex component by comparison. For it to be able to perform meaningful calculations, a relatively large number of TFTs need to be integrated.

Before Arm and PragmatIC presented their processor, the most complex metal-oxide TFT IC featured 1,400 gates. Unlike a microprocessor, however, this chip was designed for a specific machine learning task. Supporting an extended set of instructions, a processor is much more versatile. It can be programmed for a variety of applications, including machine learning algorithms.

To manufacture flexible processors, researchers also explored thinning down silicon dies before transferring them onto flexible substrates and low-temperature polysilicon technology. These methods suffer from several disadvantages, however, such as poor scalability, difficulties with system integration and high costs.


The operating principle of metal-oxide TFTs is similar to that of traditional MOSFETs, though the designs differ and different materials are used. Crucially, when making metal-oxide TFTs, the substrate doesn’t become part of the transistor. They can therefore be fabricated on a range of (insulating) substrates.

This includes flexible ones, provided processing takes place at low enough temperatures to prevent damage to the substrate. Depending on the manufacturing recipe and substrate characteristics, the resulting metal-oxide circuits can be bent to varying degrees. Some can even be rolled up tightly without getting damaged.

Metal-oxide TFT ICs are considerably cheaper to manufacture than their silicon counterparts, perhaps by as much as a factor of ten. Performance-wise they’ll never catch up to silicon, but they don’t have to. For plenty of applications, low cost and/or flexibility are the keys to success.

Some obstacles still have to be overcome before flexible processors are to hit the market. The most important one is energy efficiency. Arm’s and PragmatIC’s processor is an NMOS chip, meaning every transistor draws power whenever turned on. The power requirements of the current chip aren’t excessive, but the inefficiency becomes a problem when scaling up.

CMOS, which dissipates power only when switching, would be much more energy efficient, but engineering the complementary p-type transistors with metal-oxide materials is challenging.