Feb. 5, 2008 – With flash process technology scaling approaching its limits, along comes Nanochip, a developer of MEMS silicon data storage chips, with a technology that does not use lithography in its manufacture. The company recently completed (Jan. 22) a $14M financing round supported in part by Intel Capital and JK&B Capital, and aims to complete development of its first prototypes later this year, start design verification testing and limited customer sampling in 2009, and begin production in 2010, CEO Gordon Knight told WaferNEWS.
The company’s first products are expected to exceed 100GB/chipset, reaching terabytes in the future. Because Nanochip’s MEMS-based probe storage technology is mechanical in nature, i.e., with moving parts, its access times are in the millisecond range — too slow for programmed storage. Instead, the company is targeting NAND and disk drives. “We’re going after large database storage, i.e., servers, laptops, USB drives, and eventually, cell phones, when we get our costs down,” explained Knight.
Currently, Nanochip has four fabs around the world working on the prototype project (two partners in the MEMS sector and two in storage media), and talks are ongoing with potential production partners. While not able to identify partners, Knight did note that the MEMS-based storage chip must be produced on 200mm wafers.
The MEMS chip itself is an ~150mm2 die based on a 3-wafer construction (see figure): a CMOS substrate containing all the electronics to control the chip and provide error correction, a substrate that contains the suspended media moving platform; and a sealing cap wafer. Each die has ~8000 cantilevers and AFM probe tips, with a single AFM probe tip on the end of each cantilever. The probe tips are grown on top of the CMOS substrate and can be raised and lowered onto the media platform, which moves in raster fashion in the X-Y directions. The media layer (i.e., recording layer) is continuous, and not patterned.
Structure of Nanochip’s MEMS-based advanced memory device.
Because the media moves under suspension, there is no contact, and hence no wearout of the media, according to Knight. The probe tips, however, do contact the media. A servo controller moves the tips down into contact with the media as necessary to complete read/write operations, typically, only several hundred tips are moved at a time. Knight maintains that proper engineering of the tip material and the media surface chemistry has enabled tip life to be extended to allow very high operational usage over many years. “The media does cycle more than 10 million times and we are still testing its ultimate lifetime,” he said. “The read/write/erase rate is similar to that of a disk drive in that direct overwrite is used, thus eliminating block erase issues that flash has.”
Capacitance position sensors measure coarse position to <1nm accuracy in the X and Y directions. Data is also fine-tracked on left/right servo marks. Tip position can be corrected -- for each tip -- by correcting for pressure, temperature, and elevation. According to Knight, fine tracking accuracy is at ~0.1nm. Each cantilever in the structure can be moved to compensate for temperature or other environmental effects.
The servo technology itself is not new — “IBM did something similar on its Millipede program,” Knight acknowledged. “But we had a breakthrough a couple of years ago in a new class of media that doesn’t wear out … that [was] the stimulus that got us into full product development.” He declined to identify that media material, and indicated he probably won’t until the company is in production. He noted, however, that “the manufacturing technology does not depend on advanced lithography, and never will. We use fully written-off fabs, and all of our parts are several microns [in terms of size].” — D.V.