Static information storage and retrieval – Read only systems – Semiconductive
Reexamination Certificate
1999-12-22
2001-02-06
Nelms, David (Department: 2818)
Static information storage and retrieval
Read only systems
Semiconductive
C365S051000, C365S105000, C365S063000
Reexamination Certificate
active
06185122
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
This invention relates to low cost, high density semiconductor memories and, in particular, to semiconductor memories whose contents are “nonvolatile”: data stored in the memory is not lost or altered when electrical power is removed.
2. Background of the Invention
There is an ever-increasing demand for ever-denser semiconductor memories, and customers continue to purchase these memories in ever-greater quantities, even as the number of bits per chip quadruples (approximately) every three years. Increasingly higher densities are required, at ever lower costs, to meet the needs of the marketplace.
Semiconductor nonvolatile memories may be divided into two categories: (1) those in which data is permanently written during the manufacturing process and whose contents cannot be subsequently changed, called “mask ROMs” or “factory programmed ROMs”; (2) those in which data may be supplied after the finished memory device leaves the factory. This latter category is called “field programmable memories” because their contents may be written, by the user, when the semiconductor memory chip is deployed to its final application, “in the field”.
Field programmable memories are further subdivided into “write once” memories and “write/erase/rewrite” memories. Those written once are referred to as “PROM” (programmable read only memories) or “OTP ROM” (one time programmable read only memories). And those memories that provide write/erase/rewrite capabilities have been referred to as “UVEPROM” (ultraviolet erasable programmable read only memories) or “EEPROM” (electrically erasable programmable read only memories) or “Flash EEPROM” (fast and flexible EEPROMs). In contrast, the contents of mask ROMs are permanently stored during manufacture, therefore mask ROMs are not erasable and are effectively “write only once, at the factory” memories.
Field programmable memories are much more flexible than mask ROMs, since they allow system product makers to inventory a single general part-type for many applications, and to personalize (program the memory contents of) this one part-type in numerous different ways, much later in the system product flow. This flexibility lets system manufacturers more easily adapt to fluctuations in demand among different system products, and to update or revise system products without the expense of scrapping (discarding) existing inventories of pre-programmed mask ROMs. But this flexibility has a cost: field programmable memories generally achieve lower densities (fewer bits per chip) and higher cost (larger price per bit) than mask ROMs. Customers would prefer to buy something that offered the flexibility and convenience of a field programmable memory, while achieving the cost and density of a mask ROM. Unfortunately, such a device has yet not been available.
There are two reasons why mask ROMs have been denser and cheaper than field programmable memories. First, since mask ROMs do not support erase or rewrite functions, their peripheral circuits need not contain any dedicated circuitry or I/O terminals for input-data steering, for write timing, or for write control. Thus the peripheral circuits of a mask ROM may be smaller than those of a field programmable nonvolatile memory. This reduces the die size of a mask ROM, compared to the die size of a field programmable nonvolatile memory, allowing more mask ROM chips to fit on a semiconductor wafer, which lowers costs.
Second, since mask ROMs are written only at the factory, their memory cells may be designed and optimized for read operations exclusively, and generally their memory cells consist of only a single circuit element (e.g. a single MOS transistor). But the memory cell of a field programmable nonvolatile memory must include support for write operations. Therefore, field programmable memory cells generally contain more than one circuit element: generally a second tunnel oxide floating gate, or a write/erase series transistor, is added to the single MOS transistor needed for reading. The extra element(s) in the field programmable cell consume additional silicon area, making the memory cell area larger than the area of a mask ROM memory cell. Thus the density of field programmable nonvolatile memories has been lower than the density of mask ROMs.
Field programmable memories having write/erase/rewrite capabilities offer yet more flexibility. They permit product upgrades, field reconfiguration, and enable a host of new applications such as digital photography, solid state disks, et cetera. Unfortunately, these devices have generally suffered from lower density and higher cost than one-time programmable memories.
Turning now to the design of the memory cell used in these memories, most nonvolatile memory cells have employed semiconductor devices such as MOS field-effect transistors, junction transistors, or junction diodes, built in a planar monocrystalline semiconductor substrate. This approach allows only very limited integration vertically into the third dimension (i.e. perpendicular to the plane of the substrate), since each memory cell contains some elements built in the substrate.
Conventional nonvolatile memory cells are manufactured using a number of sequential photolithographic steps, which define the geometric shapes of the cell features. For example, fabrication of the prior art mask ROM cell shown in
FIG. 1
requires at least five photolithographic masking steps: (a) nitride-LOCOS patterning; (b) polysilicon gate patterning; (c) contact patterning; (d) metal patterning; (e) programming with ion implant patterning. These steps are performed sequentially, and care is taken to align each subsequent layer to earlier layer(s) already patterned on the memory circuit, to ensure that the geometric features of each layer will be printed in their desired spatial locations. For example, in the cell
10
of
FIG. 1
the ion implant layer would conventionally be aligned to the polysilicon layer, which was patterned previously.
Unfortunately, photolithography machines used in high volume semiconductor manufacturing do not perform these alignments perfectly. They have a “layer misalignment tolerance” specification which expresses the alignment error that may result when aligning a new layer to a previously existing layer on the memory circuit. These misalignment tolerances force memory cell designers to use larger feature sizes than otherwise would be necessary if alignment errors were negligible.
For example, if a certain feature on the metal layer were required to completely overlap a feature on the contact layer, the geometric overlap between these two features would have to be designed at least as large as the misalignment tolerance between the contact layer and the metal layer. For another example, if a certain feature on the polysilicon gate layer were required to avoid and not touch a feature on the LOCOS layer, the geometric spacing between these two features would have to be increased to be at least as large as the misalignment tolerance between the polysilicon gate layer and the LOCOS layer.
Memory cell sizes are enlarged by these misalignment tolerances, which increase die size, decrease density, and increase cost. If a new memory cell structure could be found which required fewer sequential photolithographic steps, this cell would include fewer misalignment tolerances in its feature sizes, and it could be made smaller than a cell with more photolithographic steps.
And if a new memory cell structure could be found which had no alignment requirements at all (a “selfaligned” cell), in either the X- or Y-directions, it would not need to include any alignment tolerances in its feature sizes. The new cell could be made smaller than a corresponding non-selfaligned memory cell.
FIG. 1
depicts a very popular circuit design used in mask ROMs. It is an example of the “virtual ground” class of ROM circuits as taught, for example, in U.S. Pat. No. 4,281,397. Its memory cell such as cell
10
, consists of a single MOS transistor built in the planar semiconductor substrat
Cleeves James M.
Farmwald P. Michael
Johnson Mark G.
Lee Thomas H.
Subramanian Vivek
Blakely , Sokoloff, Taylor & Zafman LLP
Le Thong
Matrix Semiconductor Inc.
Nelms David
LandOfFree
Vertically stacked field programmable nonvolatile memory and... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Vertically stacked field programmable nonvolatile memory and..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Vertically stacked field programmable nonvolatile memory and... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2607335