This article is more than 1 year old
Will it bend? That is the question: Arm boffins boast of first flexible 32-bit chip
Plastic-backed TFT part positioned as perfect for the Internet of Things
Researchers at chip designer Arm have shown off a prototype microprocessor - dubbed PlasticARM - built on flexible plastic, letting it curve around surfaces and even flex backwards and forwards.
Fabricated using a combination of metal-oxide, thin-film transistors (TFTs) and a flexible plastic substrate, offered commercially by PragmatIC as FlexLogIC, the PlasticARM is more or less a fully functional implementation of Arm's Cortex-M0+ core. The difference: you can bend it without the cracking sound you'd get from a traditional silicon chip.
Flexible circuits aren't new, but Arm claims PlasticARM is a breakthrough: a working 32-bit microprocessor boasting around 12 times the logic gates of its nearest competition. The only snag: all it can do, at the moment, is run through a test program burned into its read-only memory.
"The current ROM implementation does not allow changing or updating of the program code after fabrication," Arm's researchers admitted in their paper, published today in Nature, "although this would be possible in future implementations (for example, via programmable ROM)."
The company positioned the research as key for the Internet of Things (IoT), the ever-growing march to put digital smarts into a range of devices. PlasticARM-based microcontrollers, the company claimed, can be produced at a very low cost and conform to curved or uneven surfaces – though it has yet to prove the latter, having conducted all testing with the prototype firmly attached to a very flat piece of glass.
When asked about this, James Myers, distinguished engineer at Arm and one of the paper's co-authors, told The Reg the team was "planning to perform the bendability tests of the flexible ICs while they are being flexed. Having said that, PragmatIC has already shown that their metal-oxide transistors can bend down to a radius of 5mm."
- Nvidia launches Cambridge-1, UK's most powerful supercomputer, in Arm's neighbourhood
- Chinese chip designers hope to topple Arm's Cortex-A76 with XiangShan RISC-V design
- Sing a song of Office, a pocketful of why: ARM64 version running in a Pi
- Arm chief hits out at 'ill-informed speculation' over proposed Nvidia buyout
When quizzed about use cases for flexible processors, Myers cited examples including "smart patches for wound care and healthcare applications, identification and provenance tracking, and safety or freshness indicators for fast moving consumer goods, or in waste management and packaging recycling."
He added: "Although PlasticArm performs a fixed functionality where the program is burned into the ROM, a programmable ROM can potentially be used where the user can burn the program into the ROM in field."
What Arm has created isn't yet competition for its more traditional – and brittle – silicon chips, however. The PlasticARM functional prototype was produced on a 0.8µm process node, or 800nm, while traditional Cortex-M0+ parts are typically made at 40nm – offering a feature size, and corresponding reduction in footprint, some 20 times smaller.
The part is also clocked at just 20kHz, with the theoretical limit of its capabilities placed at 40kHz when running at 4.5V instead of 3V. The Cortex-M0+ cores in Raspberry Pi's recently launched RP2040 system-on-chip, by contrast, are officially rated to a top speed of 133MHz – 3,325 times faster – and can reach considerably higher if operated outside spec.
There's a limit to how complex PlasticARM parts could get, too – in their current form, at least. "We anticipate that the lower-power cell libraries we are developing will support increased complexity, up to about 100,000 gates," the researchers noted, representing a fivefold increase over the prototype's gate count. "Moving to more than 1,000,000 gates will probably require complementary metal-oxide-semiconductor (CMOS) technology."
Nevertheless, the team is confident that bendy processors have a place. "The aim of the TFT technology is not to replace silicon," the researchers explained. "As both technologies continue to evolve, it is likely that silicon will maintain advantages in terms of performance, density and power efficiency. However, TFTs enable electronic products with novel form factors and at cost points unachievable with silicon, thereby vastly expanding the range of potential applications.
"We envisage that PlasticARM will pioneer the development of low-cost, fully flexible smart integrated systems to enable an 'internet of everything' consisting of the integration of more than a trillion inanimate objects over the next decade into the digital world. Having an ultra-thin, conformable, low-cost, natively flexible microprocessor for everyday objects will unravel innovations leading to a variety of research and business opportunities."
Myers remarked that PlasticArm "is an embedded microcontroller (ARM-v6M CPU with modest memory, peripherals and IO) with a fixed functionality. Currently, the FlexIC technology uses NMOS logic, and it will be possible to develop more complicated flexible processors when CMOS becomes possible in this process."
"However," he added, "the primary driver for the technology is not to make increasingly complex circuitry but to add compute onto products in areas such as fast-moving consumer goods and in healthcare applications such as wound management."
Myers also noted that the tech "scales down to smaller geometries similar to silicon. Currently the process technology is at 800nm and as the process matures, the geometries of the devices will shrink.
"The flexible IC technology is orthogonal to silicon and has different drivers. This research demonstrates that it is possible to embed processor technology where silicon isn’t a viable choice, such as when cost is a primary driver."
He added: "For example, it will be too expensive to embed a silicon microcontroller into a label or package that costs only 10 cents. The key market areas for PlasticArm are sectors where form factor (thinness and flexibility) and cost are the highest priority. Whilst the technology will shrink further in the future, a 10nm process node would be uneconomic for the target markets, where a small amount of compute is needed and high density, high complexity circuitry is not required."
PragmatIC told The Reg: "One of the great advantages of PragmatIC's technology is the cost, both upfront and per unit. In terms of upfront cost the NRE (non-recurring engineering) cost is orders of magnitude lower than conventional silicon.
"The simplicity of the process, with only 13 material layers, means that the turnaround time is on the order of days rather than months."
PragmatIC added that it uses conventional manufacturing equipment but "has stripped out high-cost processes and steps," focussing "on achieving a low-cost implementation of the level of compute required for the applications being targeted. This leads to different choices than those in the silicon industry which is focussed more on very high density and very high compute capability."
When we asked whether Arm had a roadmap to commercialisation for the technology, Myers responded that as the output of joint research by Arm's boffins and PragmatIC Semiconductor, PlasticArm "is a proof-of-concept prototype rather than a commercial product, which requires more prototyping, testing and qualification."
The prototype part measures in at 59.2mm², has 28 pins, and a total of 56,340 devices – made up of 39,157 TFTs and 17,183 resistors – giving it an equivalent NAND2 gate count of 18,334. At its rated 20kHz clock speed, it draws 21mW giving it a power density of 0.4mW mm⁻².
"We do have a research roadmap to continue exploring sustainable applications of novel compute enabled by this technology. We're excited to see how and where this technology could be creatively adopted throughout the ecosystem."
Their paper, "A natively flexible 32-bit Arm microprocessor" has been published in the journal Nature. ®