Much of the focus in the microcontroller and microprocessor markets is on the development of high performance devices with low power consumption. Capable, efficient microcontrollers are being applied in greater numbers in range of industrial systems, while their microprocessor cousins are powering the latest pcs, netbooks and servers.
Yet there are applications at completely the other end of the performance spectrum, where clock speeds can be measured in Hz, rather than GHz. And it is these markets that are the focus of a programme to develop organic microprocessors. The work has been performed through a collaboration between Belgian research centre imec and Dutch R&D organisation TNO within the Holst Centre framework.
The work picks up on earlier programmes addressing low cost rfid tags. Heading the project is Jan Genoe, leader of imec’s organic and polymer electronics group. “Imec has been working on rfid tags for some time,” he noted, “with the aim of being able to print them on any product. They are cheap circuits which can be read when close to an antenna.” He said the tags featured 1200 transistors and could communicate at 2kbit/s at 13.56MHz. “They were slow, but good enough to send a product code.”
Now, the focus has shifted to determine whether it is possible to create a microprocessor using the same basic technology. And imec has demonstrated an 8bit microprocessor with 4000 transistors that runs at 6Hz.
Genoe said organic electronics would always be slower than silicon because of the mobility of organic molecules. “It’s 1000 times less than silicon,” he noted. “And gate lengths can’t be scaled because the devices need to be bendable.
Where gate lengths are much less than 100nm in silicon, they are no shorter than 5µm in organic electronics.” Genoe is keen to point out the similarities between imec’s 8bit microprocessor and Intel’s first such device, the 4004, introduced almost 40 years ago. “We are working on a p-type device and the 4004 was also a p-type device. And where the 4004 needed a 15V supply, our device can operate from a 10V supply.
We have used the 4004 as a comparison to show that we can make an 8bit device with all the instructions needed.” But he emphasises the 6Hz clock speed. “We can’t get speed because of the low mobility and the gate length; it’s like being back in the 1970s. But we haven’t aimed to match the performance of modern electronics and it isn’t possible.” The project is using technology developed by Philips spin out Polymer Vision to make the microprocessor.
This uses pentacene – a molecule based on five benzene rings – as the basis for its approach. Genoe explained why: “The pentacene molecule has a lot of orbitals that overlap, so we can get good transport from one molecule to another.” While the ideal mobility in a device is 1cm2/Vs, imec prefers a lower mobility, but one which is uniform across the wafer. “That gives more reliability,” Genoe asserted. “A few years ago, we would have gone for maximum performance; now, it’s a different approach. Knowing how to get to maximum performance is important, but we are focusing on reliability.”
Organic thin film transistors
Polymer Vision’s technology was originally developed as the basis for rollable displays. In that application, the top gate was the pixel electrode in the display’s backplane. However, for the microprocessor, that gate is used as a back gate to control threshold voltage, Vt. The project has developed a way to create dual Vt logic using organic thin film transistors. In the Polymer Vision approach, the top gate acts as the pixel electrode in the display backplane.
However, imec has used this as a back gate to control independently the Vt of drive and load transistors in a basic inverter. The microprocessor has been produced on a 25µm thick foil substrate. Unlike attempts by other researchers, which used a stamping approach, this device is built on the substrate. “The foil is very thin,” said Genoe, “and will roll up if left on its own, so we have ‘glued’ it to a silicon wafer substrate for handling purposes.
The manufacturing process is to lay down the gate, then the gate insulator, followed by the source-drain. The organic semiconductor is followed by an insulator and the second gate. “Some of the circuitry is deposited by spin coating, with metal laid down by evaporation,” Genoe said. “It’s a simple process, with a limited number of steps. We have three metal layers, compared to the nine in silicon.” Once the device is completed, it can be peeled off the support wafer.
The processor’s core is an 8bit alu, which implements classical logic (AND, OR, NOT), arithmetic (add, subtract, increment, decrement) and shift (logic shift left, arithmetic shift right) operations. The alu is controlled by the three least significant bits among the 10bits from the processor’s opcode. The alu’s output is stored in the accumulator register, with three further working registers and an output register available.
Genoe’s team has also developed an instruction foil which can be connected to the microprocessor. In the demonstrator, this foil enables a running averager program which can process one input per second. According to Genoe, this is a typical digital signal processing instruction and the processing of sensor outputs is a likely application for foil microprocessors. One of the development goals for the project is to improve spread: the relationship between Vt and mobility.
“This varies, depending upon process conditions,” Genoe noted. “But we want the Gaussian distribution of Vt to be as narrow as possible.” Asked why the device was designed to use a 10V supply, Genoe said the supply voltage can be up to five times the Vt in a p-type semiconductor. So, Vt could be 2V from the 10V supply. “But the Gaussian distribution around a Vt of 2V means a higher probability of transistors with a Vt of less than zero. This means we can integrate fewer transistors.”
The microprocessor foil comprises 3381 transistors and measures 1.96×1.96cm, while the instruction foil features 612 transistors and measures 0.72×0.64cm. The power consumption is said to be 92µW when running from a 10V supply. Genoe believes organic processors will never be fast. “But, using the knowledge we’ve gained, I think we should be able to get at least an order of magnitude improvement and that’s on the road map.”
In terms of applications, Genoe envisages printing a microprocessor on a cereal box. “This small processor could be accompanied by, say, four pushbuttons. If you take cereal out of the box, you can push the buttons and see how many calories that represents on an oled display. Or you could print a small game on the box, with customers buying a particular product because of the game.”
And such systems could be powered by a microbattery generating a higher voltage, but very low current. Nevertheless, the production issues have yet to be addressed. “We’ve shown that a microprocessor can be printed,” he concluded, “but it hasn’t been done in volume yet.”