Low-Cost Embedded NVM for Power Management Designs

by Yakov Roizin, Evgeny Pikhay, Amos Fenigstein, Avi Strum, Tower Semiconductor Ltd., P.O.Box.619, Migdal HaEmek 32105, Israel
ec810smh100aec810smh100cIn this paper we present two unique NVM solutions which can close two significant gaps in the embedded  memory  IC world. Today’s designers have a large choice of  embedded non-volatile memories (NVM) verified in different production technologies. These memories can be divided into three groups:

* true MTP (multiple-time-programmable)  having densities up to several tens of Mbit and  endurance of >100 K program/erase cycles [1,2] 
* OTPs (one-time-programmable) fabricated without additional masks, enabling NVM modules up to 1 Mbit [3]; 
ec810smh100d* small size (up to several tens Kbit) MTP or FTP (few-times-programmable) without mask adders [4,5].

ec810smh100bTrue MTP  usually requires over 10 additional masks to the core CMOS.   OTP modules can not be used efficiently for memory sizes larger than 1 Mbit due to reliability limitations,  and also are limited in field  applications. The no added mask MTP solutions are area consuming and usually limited to no more than a few tens of Kbits.  Cost-efficient MTP solutions for middle-sized (<1Mbit)  and  large (>1Mbit) embedded memories are not available.
In the design of power management (PM) ICs, similar to other application fields, embedded NVM at no additional cost is one of the main value-added differentiators. Both small-size NVM for trimming of analog parameters, feature selection, address storage, etc., and larger NVM modules for code storage, system configuration and security applications are challenging.  Nevertheless, to find a low-cost embedded NVM for a typical Fig1.bmp0.18/0.13um PM process flow (technology nodes compromising production costs and availability of cutting-edge IP libraries [4])  is  much more difficult than for standard logic applications.

Compared with standard logics, PM chips  are typically designed for a more robust  5V only or 5V/1.8V operation, even if larger silicon area is required. Thicker GOX in the  5V designs implies high operation voltages, thus  requiring  area consuming charge pumps, complicating field programming and  introducing additional reliability limitations in the scaled down core CMOS . The known MTP NVM solutions suitable for PM IC have very large cell areas, thus adding significant die area even for small size embedded modules. OTP NVM solutions do not allow field upgrades and do not support 5V I/O voltage typically used in power management ICs.     
In this paper we report on two novel types of low-cost memories (Y-Flash and S-Flash) suitable for PM,  both having record cell sizes in their application fields. The memories are highly scalable and can be embedded into typical CMOS process flows (specifically , but not limited to 0.18um and below) without additional masks or with up to two additional non-critical masks (for memory performance enhancement).
Y-Flash is targeted for 1Bit to 1Mbit memory modules.  It has an original “asymmetric” two-terminal cell design with only ~2 um2/Bit, and  allows more than 100,000 program/erase cycles. Y-Flash is supposed to close the mentioned  “low-cost middle-sized MTP” gap in the NVM market.  Fig.1 shows the Y-Flash operating principle and a layout of a typical Y-Flash cell. The FG has a Fig2.bmpportion 121 disposed over the channel region and an additional portion 122 extending into an enlarged drain diffusion area away from the channel region.

The gate-to-drain capacitance CDS is thus significantly higher than the gate-to-source capacitance CGS.  The width of the FG extension portion 122 is minimized, enabling LDD implant merging under the floating gate extension.  The HV LDD implant in the NVM transistor is substituted by other implants from the core CMOS process. Programming is performed  by applying a positive programming voltage to the drain region and connecting the source region to ground, thereby causing channel hot electron (CHE) injection from the drain region into the FG.  In the  low-power flavor,  the memory cell  is placed into an insulated p-well to allow programming with more efficient CHISEL  (secondary electrons) mechanism that assumes substrate  bias.  

Erasing the floating gate involves connecting the drain diffusion to ground and applying the positive voltage to the source, thereby causing band-to-band tunneling (BBT) generation of holes and their injection into the FG. There is no need for a select transistor in the Y–Flash designs. This is Fig3.bmpillustrated in Fig.2 where a simplified  Y-Flash cross-wise memory array is shown. The row and column select circuitry is built using   LDMOS transistors which are legal devices in the same core CMOS process flow (no additional masks) or by  cascading solutions.  

Reliable  studies  confirm feasibility of over several hundred thousand  Y-Flash program/erase cycles.  A very small cell area (compared with competing single Poly FG) allows MTP modules significantly smaller than  in typical MTP designs (which are in the range of 10 ?m2, 5x larger than the Y-flash cell).  Accordingly, asymmetric Y-Flash are ideal for incorporating into low cost CMOS integrated circuits requiring several Kbits uo to 1Mbit  of high endurance memory.
In plenty  of  high density (>1Mbit) embedded applications there is no need for true MTP. A typical example is a microcontroller code which is not changed too many times.  In such cases, it would be favorable to replace the designs currently using true MTP large volume embedded flash with low cost solution. Though having a record memory cell size among single Poly FG memories, Y-Flash  becomes too area consuming for modules  exceeding 1Mbit.  Instead,  S-Flash is targeted for the use in applications requiring cost-effective high density (1Mb-32Mb) FTP embedded memories. S-Flash  employs  Tower patented spacer trapping principle[6] and has an extremely small cell size (<0.2um2/Bit, close to the well known Tower’s microFlash® and Saifun’s NROM[2]). The principle of S-Flash is illustrated in Fig. 3.a.

In a standard CMOS manufacturing process, the nitride of spacers  is  deposited on a thick (~200A) CVD oxide. Decreasing the  bottom oxide (BOX) and sidewall oxide thickness, skipping of LDD extensions and performing special drain engineering to enhance lateral fields in the drain region,  facilitates trapping of electrons in the nitride. A standard n-channel MOS transistor is thus converted into a non-volatile memory cell. The trapped in the spacer electron charge is monitored in a reverse read-out scheme [2] .Two bits of information (in the spacers from both sides of the gate) are stored in a single memory transistor, similar to NROM memories.

The cell can be erased by hot holes, generated by the BBT mechanism like in NROM and Y-Flash. Memory windows over 2V and small programming times were confirmed in measurements of the developed 2MBit Array-TEG. Area saving H-array geometry [6] is employed in the existing designs (Fig.3.b). The S-Flash memory allowed reliable operation up to ~ 100 program/erase cycles. Programming and erase  with voltages below |5|V, excluded the need of special high voltage periphery circuitry. No disturb issues different from those known for NROM memories were revealed in the studied range of cycles and employed voltages. The high volume S-Flash memory  is embedded into the standard CMOS process with only two non-critical masks added and minimum additional process operations.
In summary, two original Tower NVM technologies allow the design of low-cost memory modules covering all practically necessary sizes of NVM in embedded PM applications. A record for single Poly MTP memory cell size and simple array design  with no mask adders  makes, in some cases, the single Poly  Y-Flash as effective as the true MTP Flash, with no additional cost, typically associated with a true MTP solution.  S-Flash featuring  one of the smallest in the industry cell size, is the cheapest FTP large volume memory solution not requesting special  periphery devices and  having outstanding retention performance.

2. B. Eitan,  IEEE Electron Device Letters, v.21 , 543, 2000.
5. B. Wang and Y. Ma, NVSMW-ICMTD08, 2008 pp.22-26
6. Y.Roizin and A.Fenigstein, US Patent 7,227,234, 2007

Prof. Dr. Yakov Roizin ,  Senior Staff Member and NVM R&D Manager at Tower, has more than 30 years of semiconductor device development  experience. 

Evgeny Pikhay, Senior Device Engineer at Tower, graduated from the Technion-Israel Institute of Technology and has 7 years of experience in NVM development.

Dr. Amos Fenigstein, R&D Director at Tower,  received his degrees form Technion-Israel Institute of Technology and has  over than 15 years of semiconductor-industry experience in engineering and  management.

Dr. Avi Strum is a Vice President and General Manger of Tower Specialty Product Line.  Prior to joining Tower  he directed R&D activities and held strategic management positions at several semiconductor companies including Intel Corporation. Avi’s email address is: