ChipCrunch.com

While on the topic of novel memories, we might as well continue down this path, this time with a transistor-less memory cell designed for non-volatile memory products. Before delving into the memory cell details and the associated physics, a little info about the company, namely Unity Semiconductor, might be of interest. Based out of Sunnyvale, CA and founded in 2002, Unity Semiconductor emerged from stealth mode just about a month ago as it announced that it raised additional $22 million in Series C funding. This brings to total amount of funding that the company has received to a very respectable $75 million over the last few years, proving that an interesting idea in the semiconductor space will attract some serious money. The company was founded by Darrell Rinerson, Ed Ward and David Bostwick, all of whom seem to have had stints at some of the major semiconductor companies, including Micron and AMD. It is fairly unusual for a company to wait this long before existing stealth-mode, but just like with the device physics that took time to calculate, Unity Semi has also been calculative on how to penetrate the market and maintain control of their new conductive metal oxide and ionic motion (CMOx) technology. For this purpose, the company has amassed a war chest of patents before existing stealth-mode. How Many? How about in excess of 60 granted and another 90 applications currently pending, so feel free to knock yourself out examining them.

As mentioned before, the technology is targeted at the non-volatile memory (NVM) market which the company beliefs will grow to more than $25 billion by 2013. Currently, this market is being served by NAND flash technology; however there is a concern in the industry that NAND technology scaling might be reaching its limits. Hence, the recent spur in the development of new technologies that are to replace NAND flash including: resistive ram (RRAM), ferroelectric RAM (FeRAM), phase-change memory (PCM), magneto-resistive ram (MRAM), multi-chip packaging (MCP), and finally 3D-memory. Unity Semi briefly discusses the cons for each of these technologies in section 7 of this document. The company intends to tape-out a 64-gigabit product in the first half of 2010 and ramp up to volume production in the first half of 2011.

The figure above shows a cross section of the CMOx technology. On a high level, through the application of a high electric field across the tunnel oxide (TO) layer, the charge buildup in the tunnel oxide can be controlled which leads to a change in the trap-assisted tunnel current through the oxide, or in other words a change in the device resistance. If that just went over your head, you’re not the only one for sure; thankfully Unity Semiconductor has a presentation that discusses all of this in detail, including oxygen ion mobility and the associated memory effect. This is a very good presentation indeed and a detailed paper is included as well, so it is well worth your time. The company claims that the current difference between the two resistive states is approximately 10x and that the program and erase times are in the 1us range at +/-3V. Compared to flash NAND technology, Unity Semi is claiming a 4x density improvement and a 5 – 10x write speed improvement.

Overall, Unity’s technology is interesting and impressive at the same time, which explains why the company has been able to raise this much capital. The BEOL approach to commercialization, where the memory layer with CMOx technology is deposited on top of a regular CMOS logic layer by a secondary fab seems possible as well, although it will require a good partnership with an integrated device manufacturer (IDM). At the same time, NAND flash technology should not be written off too quickly. For years, people have predicted the end of transistor scaling due to various reasons, and yet the scaling continues and if you talk any of the major semiconductor players you will find them to be optimistic about being able to continue this trend for several process nodes to come. From published material it seems that the program vs. erase current is predicted to scale with device size, but it will be interesting to observe whether Unity Semi can maintain the 10x difference, especially as the technology is moved to next process nodes. On the flip-side, because of the BEOL approach, the company does not need to necessarily move to the latest and greatest process, since the CMOS logic layer is able to use a different process than the CMOx memory layer, which is a very nice option to have.

In April we discussed Nethra’s acquisition of Amric’s assets and what the future might hold for the products Ambric was developing. Shortly afterwards I received a ping from Nethra’s Senior Director of Business Development, Manu Pallai, to check their website in about a week for some updates. I have to admit I sort of dropped the ball on that one, and only recently got a chance to follow up with what Manu was alluding to. As it turns out, without as much as a press release, Nethra has now made available a whole slew of Ambric based development boards: the Am2045 GT, the AM2045 GT2, the Am2045 IDB, and finally the Am2045 GT2-SDI.

As the names imply, all of these boards are powered by the Ambric’s massively parallel processor arrays. All of these boards feature a PCIe interface and can be plugged directly into a regular system for easy development. The difference between the GT and GT2 is that the latter offers two Am2045 processors. The Am2045 IDB is a little bit more versatile and can be configured with a choice of processor arrays: Am2045, Am2029, and Am2016. It contains four 32-bit GPIO ports and an FPGA for custom logic. It also features a Serial Flash, a USB interface, and an ATX connector to expand the configuration options. The GT2-SDI is a reference design which adds a whole bunch of digital connectors and stream operations. The programming environment for these boards is the aDesigner Tool Suite and integrates Ambric’s structured object programming model (SOPM). We discussed the Am2045 and SOPM in more detail back in 2007 in this post. Nethra does not provide any pricing info for these development boards on their website at this time, but if they are reasonably priced I might have to pick one up just to have something this massively parallel to play with and possibly pit against a general purpose CPU.

Back in March we covered TimeFlux technology from LightWaves. The article sparked quite a discussion with people referencing the Shannon-Hartley theorem, time-frequency duality, and so on while discussing the performance claims made by LightWaves. The discussion focused mainly on wired networking and the ability to provide proper bandwidth and latency for real-time data streaming. Assuming that all this data in some way or another arrives at a particular location, be it an enterprise LAN or a home network, a good way to distribute it to all the clients will still be needed. With UWB technology floundering as of later, it seems that an alternate wireless distribution method will be needed.

This is precisely the market Quantenna is targeting: reliable high-speed wireless coverage for high definition (HD) video and related multimedia content. Based out of Freemont, CA the company recently raised $14 million in Series C funding from Southern Cross Venture Partners, Grazia Equity Gmbh, Sequoia Capital, Sigma Partners, and Venrock Associates. Founded back in 2006, Quantenna has raised a very respectable $42 million in venture funding to date. The Series C funding is mostly to be used for expanding sales and engineering activities. Currently, the company offers free flavors of their 802.11n Wi-Fi compliant chipsets: the QHS1000, the QHS600, and the QHS450. The numbers in the product names correspond to the maximum link speed that each chipset is able to support. The flagship product, the QHS1000, can operate in the 2.4 and 5 GHz spectrum concurrently, while the latter two work exclusively in the 5 and 2.4 GHz spectrums, respectively. The two most touted features that these chipsets support are the 4x4 multiple input/multiple output (MIMO) technology and dynamic digital beamforming.

According to the company press release, the combination of these technologies results in a four time larger coverage area and twice the throughput of existing solutions. At the same time the product page states that these technologies provide a 50 percent performance improvement over other 802.11n solutions. I’m not a networking expert by a long shot, but if you claim twice the throughput and a four time larger coverage area, wouldn’t this indicate more than a 50 percent performance improvement? Of course, it can be that performance includes additional factors that are not accounted for in the initial statement, or maybe the baselines for the two comparisons are different, nevertheless some clarification would be appreciated. The 4x4 configuration refers to four radio transceivers and four antennas which in addition to supporting two data streams can utilize the extra two antennas for dynamic digital beamforming to increase data stream reliability by 12dB over 3x3 configurations. Optimum performance is achieved by continually monitoring real-time packet data and adjusting the wireless signals in milliseconds as needed. A short video describing beamforming can be viewed on the company’s website. Additionally, Quantenna has filed several patent applications including one for a tunable antenna system and one for interference rejection, in case you might enjoy further reading.

Some things just don’t go away easily, and in some rare cases this might actually be a good thing. We wrote about Ambric and their promising Am2045B processor and development environment back at the end of 2007. Subsequently, at the end of 2008 after failing to raise an additional $15 million in funding, Ambric folded. Now it seems that there will be an additional chapter in Ambric’s history for the company’s assets were recently acquired by Nethra, as I was recently informed by Nethra’s Senior Director of Business Development, Manu Pallai, via a note titled Ambric Continued. According to a press release on the company’s website, Nethra intends to complete the integration of the Ambric team by mid-April and continue working with current Ambric customers to ensure a seamless transition and continued support.

So who is Nethra and what use could they have for the Ambric team and assets? Let’s start with the basics: Nethra is a privately held fabless semiconductor company based out of Santa Clara, CA. The company specializes in imaging and video solutions for the digital consumer market, as well as surveillance and industrial applications. It is backed by Artiman Ventures, GunnAllen Financial, iLabs Capital, and Quicksilver Ventures. The amount of funding the company has received is presently not being published, but digging around the net it seems that the company received $14.2 million at the end of 2006. IC development costs are rather high and with this acquisition I would think that the company must have obtained some additional funding in the mean time. Nehtra’s current flagship processor is the Nethra Imaging (NI) 2065 which integrates a 32-bit RISC CPU core and supports CMOS sensors up to 3.2 megapixels. The key advantages the NI2065 offers are algorithms the company has been able to implement for adaptive lighting, low-lighting, and image stabilization to improve image quality. The NI2065 consumes a maximum of 0.2 Watts, measures 5 mm x 6 mm in a 68 pin package, and as such is targeted at mobile applications. Below is a chart summarizing the features for several of the NI 20XX processors.

Now consider the following: Nethra is heavily vested in improving the image quality captured by CMOS sensors, essentially striving to enable performance comparable to CCD sensors but at a much lower price point. Their current products max out at 3.2 megapixels and rely on algorithms to compensate for the image quality captured by the CMOS sensors. With higher image resolutions just around the corner it is likely that the 32-bit RISC core will become overwhelmed regardless of how optimized the code for the algorithms might be. The good news is that image processing is a highly parallel task where the same filter can be applied to different areas of the picture simultaneously. Now, if these algorithms can be combined with the massively parallel architecture that Ambric was pursuing, and the power can be kept reasonable, then Nethra might indeed be plotting a course that should enable it to scale well as the picture and video demands become more complex. It seems like a showdown between Nethra and Movidia might be looming in the near future.