Nano-RAM, or NRAM, includes a thin layer matrix of randomly organized CNTs often adjacent to or within a transistor stack. When a voltage pulse is applied the matrix is flexed to create many points of contact between CNTs and, as a result, multiple conduction paths through the matrix layer thereby lowering resistance. Van der Waals forces hold the CNTs in contact after the pulse. To reset the memory a reverse voltage is used and heating/phonon effects break the van der Waals forces allowing most CNTs to separate and reducing the number of conduction paths and raising the resistence. However, those van der Waals forces also make the memory highly reliable. It retains memory for more than 1,000 years at 95 degrees C and more than 10 years at 300 degrees C, Nantero claims. The resulting memory offers switching speeds of the order of 20 picoseconds at low energy together with a practical write speed of 5ns with endurance of the order of 10^11 cycles tested so far without failure, and an expectation of far higher endurance.
1T1R NRAM memory cell. Source: Nantero Inc.
Nantero uses an intellectual property licensing business model and has claimed for several years to have a number of licensees and to have its technology installed in multiple wafer fabs. But Nantero has not been at liberty to reveal many names.
So we started by asking on when the first products would come to market.
Next: Products; when?
“Fujitsu is on track for production in 2019. The Mie Fujitsu foundry will be making NRAM for Fujitsu and for other customers as well. It is a 256Mbit NRAM,” said Schmergel. What about Fujitsu moving to 40nm? “That remains the plan. But it is up to them to announce the timing of the introduction of 40nm.” Schmergel states that the Fujitsu implementation of NRAM is initially as an embedded memory.
Currently Fujitsu Semiconductor provides ferroelectric RAMs (FRAMs) and microcontrollers with FRAM on-chip although it is thought that FRAM will have difficulty in scaling to leading-edge processes, hence interest in alternative non-volatile memories.
Nantero also has at least one licensee working at 28nm on stand-alone 16Gbit DDR4 non-volatile DRAM replacement with a four-layer construction, Schmergel said. “This is much lower power, and at a lower price, than DRAM.” There is also the option to produce a two-layer 8Gbit NRAM, he added.
The power consumption save comes from both the low-energy write and the non-volatility, which allows systems to be switched off and retain data, Schmergel said. In contrast, DRAM needs to be refreshed to retain data. The proposed 16Gbit chip has a similar die area to an equivalent DRAM but as a four-layer memory it uses a more mature process technology, thereby saving cost. In the DRAM field Samsung has recently moved to a second-generation 10nm-class manufacturing – so-called 1Ynm technology – to manufacture 8Gbit DDR4 DRAMs.
However, Schmergel said he doesn’t expect to see samples for these stand-alone NRAM devices until late 2019.
For a company that has been working on its technology since 2001 and is still yet to get products into the market this could be frustrating, but as Nantero is a company with an ARM-like business model it is up to licensees to determine when products come out, said Schmergel.
One of the most important developments, Schmergel said, is a cross-point architecture that does not require a selector device. Schmergel said Nantero made the development a few years ago, has not presented it at any learned conferences but that it allows for a compact array. Schmergel gave no indication of how Nantero gets round the “sneak-path” problem in cross-point arrays that do not either use a blocking diode or selector device. Schmergel said that conventional 1T1R NRAM arrays would be used for lower density products.
Next: Long-time coming
Does the protracted time to market create a risk that a product, such as an NRAM with a DDR4 interface, could have missed its market opportunity by the time it comes out?
“No, for a couple of reasons,” said Schmergel. One is that DDR5 is only at version 0.3; the final version is not yet decided. Secondly it would be relatively simple amendment to the design to swap out a DDR4 interface for a DDR5 interface once the details are agreed. Thirdly, Schmergel asserted, there will be a long tail for DDR4 applications.
But what about taking NRAM on beyond 28nm CMOS? Is Nantero engaged with FinFET or FDSOI semiconductor manufacturing proponents, or both? “We have done a lot of work studying the scalability of our memory. In our labs we have demonstrated bit cells at 15nm by 15nm and a detailed model shows switching operation should go down beyond 5nm.”
As NRAM only requires a single mask to define where the CNT layer goes and a small number of process steps it can be fabricated at low cost and is compatible with both 3D multilayer architectures and multi-level cell operation. “One of the great things about NRAM is that it is largely process independent. We’re not affiliated with either camp – FinFET or FDSOI – but equally applicable,” said Schmergel.
Because each CNT memory bit contains 100s or 1000s of CNTs it is possible to make multi-level cells. “We have made progress with multiple bits per cell but it does end up being a lower speed memory. Still faster than flash but below DRAM, so these things are a trade-off,” Schmergel said.
Schmergel is hopeful that further licensee disclosures can be made in 2018. The company’s website states that two infrastructure systems companies, six additional semiconductor licensees and multiple leading foundries will be announced “soon.” But until those companies start selling chips there is little incentive for them to go public on their involvement with Nantero, Schmergel said.
Next: Lining up applications
In terms of applications Schmergel is hopeful that NRAM can replace battery-backed DRAM cache chips in the front end of magnetic hard drives and solid-state drives. If the DRAM is replaced with NRAM no supercapacitor is required to keep data alive and the superior performance of NRAM would improve the speed of the drive, Schmergel said.
Schmergel is also expecting that NRAM could find application in smartphones and tablet computers where it could insert as a new memory layer between DRAM and NAND flash.
For now NRAM is not going to compete with multilayer NAND flash but it could displace DRAM while offering non-volatility and thereby offering faster operation and improved standby mode. “Customers will also look at a dedicated NRAM-plus-flash architecture,” Schmergel said, although may prefer to make gradual transition for reasons of engineering management caution.
“Automotive is also a focus area for us and we are working with automotive and automotive supply companies,” said Schmergel. “The need for fast memory in automotive has skyrocketed and the high temperature operation of our memory makes it a good choice.”
Schmergel said he also expects to be able to displace flash memory in IoT applications where NRAM offers low power in both active and standby modes. “We are substantially lower than flash in write energy.”
There is also scope for using NRAMs as replacements for DRAM in articifical intelligence and machine learning applications. “Speed is an issue for these applications, power is an issue and heat is an issue.” Schemergel said that initially NRAMs deployment would likely be as DRAM replacements in data centers running AI applications. But he added that, as with other non-volatile memories such as phase-change memory (PCM), researchers are looking at how to use the CNT stack as an analog for neuronal behaviour and how to use an NRAM array of neuromorphic computing. “NRAM could perform neuronal processing at far higher speeds than PCM is capable of, at lower power and temperature stability than PCM and with higher endurance,” Schmergel concluded.
Next: More money needed
Nantero could be described as the oldest startup in town having raised more than $110 million in venture capital over its 17 year history. More recently, it announced raising $30 million in June 2015 and $21 million in December 2016, raising the possibility that more money will be required to get NRAM over the line and into production.
“Later this year we do expect to raise another round of funding. We have seven strategic investors plus a new customer with a strong desire to invest. It is natural that once a customer is on-board they want to share in the ownership of the technology.”
And, it seems, that one more year is required for NRAM to come into production. Which after seventeen years of R&D is perhaps not so long to wait.
The technical storage or access is strictly necessary for the legitimate purpose of enabling the use of a specific service explicitly requested by the subscriber or user, or for the sole purpose of carrying out the transmission of a communication over an electronic communications network.
The technical storage or access is necessary for the legitimate purpose of storing preferences that are not requested by the subscriber or user.
The technical storage or access that is used exclusively for statistical purposes.The technical storage or access that is used exclusively for anonymous statistical purposes. Without a subpoena, voluntary compliance on the part of your Internet Service Provider, or additional records from a third party, information stored or retrieved for this purpose alone cannot usually be used to identify you.
The technical storage or access is required to create user profiles to send advertising, or to track the user on a website or across several websites for similar marketing purposes.