1T OTP Memory: Delivering Quality and Reliability
By Wlodek Kurjanowicz
Founder and CTO
Sidense
Introduction
Today's consumer electronic and wireless
markets are under heavy pressure to reduce costs, increase performance,
minimize power consumption and increase security.
Configurability and programmability,
which is accomplished with one-time programmable (OTP) memory, becomes
the key enabler to achieving these goals. OTP addresses such usages
as redundancy and option select, coefficient storage, analog trimming,
RFID and chip ID, boot code, firmware and secure encryption key storage
(such as HDCP keys for HDMI).
What is needed is an area-efficient,
low-power, high-performance and highly reliable OTP memory that helps
consumer electronics and wireless manufacturers reduce costs, serves
their high density and high performance code storage requirements and
ensures that encryption keys are securely stored. The answer to this
requirement is Sidense's unique, patented, single transistor (1T)
antifuse 1T-FuseTM OTP memory, which offers higher performance,
density and reliability than any alternative NVM solution.
This paper examines 1T and 2T/1.5T (two
transistor) OTP memory cells and explains why Sidense's 1T cell is
significantly more reliable than any of the alternative multi-transistor
antifuse devices.
A Brief History of Antifuse
Antifuse technology has been employed
by the semiconductor industry for over 25 years. An antifuse is defined
as an electronic device that changes state from not conducting to conducting
or from higher resistance to lower resistance in response to electrical
stress (a programming voltage or current). Initial antifuse inventions
focused on a capacitor blown between crossing conducting lines (1969).
An MOS gate oxide breakdown antifuse was invented by Texas Instruments
in 1979.
In 1982, a dual-gate-oxide two-transistor
(2T and 1.5T) MOS antifuse was introduced, leading to what has become
a viable OTP embedded memory architecture. Since the adoption of a dual-gate-oxide
technology (3.3V/1.8V) in 0.18um technology, MOS antifuse has become
available in a standard CMOS logic process which does not require any
additional masks or processing steps.
Subsequent advancements in CMOS technology,
however, posed new challenges and barriers to the classic 1.5T antifuse
architecture: density, reliability and portability between fabs. 1.5T
memory cells do not scale well with process. Reliability is compromised
by tail bits in the programmed cell current distribution. Portability
between fabs is limited by high cell sensitivity to drain engineering.
Sidense's patented 1T-Fuse architecture
overcomes all the aforementioned issues, paving the path for reliable,
standard OTP solutions in advanced CMOS technology nodes. Sidense's
patented split-channel memory cell scales well with process, does not
suffer any dependence on drain engineering and is the only single transistor
OTP solution in the industry that does not require any additional masks
or processing steps.
1.5T Cell Reliability and Programmability Concerns
The cross section for
a classic 1.5T antifuse cell is shown below in Figure 1.
1.5T Antifuse Cell figure
A Classic 1.5T Antifuse Cell
Figure 1. A
1.5T bit cell can experience oxide breakdown (program) in three different
regions, resulting in a multi-modal distribution of programmed cell
current with tail bits compromising programmability and reliability.
As can be seen in Figure 1, programming
of the 1.5T cell is not well confined in the channel region. In fact,
the 1.5T cell can program (experience oxide breakdown) in three different
regions: (1) the channel region, (2) the leakage control (pocket or
Halo) ion implantation region, and (3) the LDD diffusion extension region.
This results in a tri-modal distribution of the programmed cell current
and is very susceptible to large tails in the current distribution.
Breakdown to the LDD region (3) forms
a resistive link between the n+ polysilicon and the n+ diffusion, resulting
in a high current tail. Breakdown to the channel region (1), which
is the dominant mechanism, forms a diode-connected NMOS transistor characterized
by a particular threshold voltage and resistance. Breakdown in
the pocket or Halo region (2), which has a higher p+ concentration,
results in higher resistance or higher Vt, leading to a low current
tail.
Tail bits compromise reliability and programmability
The tail bits in the multimodal distribution
make the program verification process extremely difficult and error
prone. The cells programmed in the Halo region (2) might be weaker than
the partially programmed or leaky cells in the LDD region (3). Since
the partially programmed cells should be eliminated, as they can heal
with time, the program and verification conditions must be experimentally
optimized for each process and each fab. This leads to a compromise
between reliability and programming yield.
Self healing of the partially programmed cells
If the read reference is adjusted to
read weakly programmed bits, on the order of 1uA, a cell that has not
yet been fully programmed but shows increased tunneling leakage current
(>1uA to LDD), which is the first stage to oxide breakdown, may be
incorrectly classified as a programmed cell. The charge, which is trapped
in the oxide and is responsible for the leakage, can eventually be removed
during device operation lifetime resulting in "self healing" and
poor device reliability since the cell will no longer be viewed as "programmed".
Higher read voltage limits retention
The low current tail
bits require higher read voltage and/or longer read time, both of which
increase exposure of the un-programmed memory cells to the elevated
voltage during every read operation. This shortens the total memory
retention time according to TDDB curves for the gate oxide.
Portability limited by the LDD process dependence
Direct exposure of the gate to the diffusion
edge in the 1.5T cell results in strong sensitivity of the cell to drain
engineering, limiting portability between fabs.
Since the LDD technology details (pocket
or Halo implantation) are manufacturing-equipment specific, the 1.5T
qualification must be repeated not only for every node, but for every
fab in a given process within the same foundry. Qualification should
also be repeated if any process improvement has been made to the NMOS
transistor.
1T OTP - Delivering Controlled and Reliable Programming
Figure 2 shows the cross section of Sidense's
patented single transistor antifuse cell (1T-Fuse), in which reliable
and repeatable programming is entirely contained to the channel region.
This novel memory bit cell scales well with process and does not suffer
any dependence on drain engineering. It is the only single transistor
OTP solution in the industry that does not require any additional masks
or processing steps.
Figure 2. The 1T-Fuse bit cell is a two-terminal, split-channel device that looks like an MOS capacitor in the un-programmed state and a diode-connected MOS transistor in the programmed state. All programming occurs in the transistor's channel region for high reliability and repeatability.
The 1T-Fuse cell uses a patented split-channel,
variable thickness gate oxide manufactured in a standard-logic CMOS
process with no additional masks or process steps. The gate oxide underneath
the polysilicon gate consists of both thick and thin oxide regions,
where the thin oxide is used for programming (breakdown). This area-efficient
and highly portable design has been verified in a wide range of process
nodes from the leading foundries including 180nm, 130nm, 110nm, 90nm
and 65nm.
Figure 3 compares the programmed cell
current distribution for 1T and 1.5T cells manufactured on the same
chip and programmed and tested under identical conditions. As can be
seen in these histograms, the cell current for the 1T cell is well controlled
and contains no tails in the distribution. This is contrasted with the
1.5T cell, which produces large tails and, as explained earlier in this
paper, leads to quality and reliability issues.
comparison of multi-modal distribution of programmed cell current for a 1.5T cell and repeatable cell current of a 1T cell graph
Figure 3.
This graph shows the multi-modal distribution of programmed cell current
for a 1.5T cell as contrasted by the
well contained, repeatable cell current of a 1T cell.
Improved Yield and Reliability
Absence of the large tails in the 1T
cell current distribution allows for better program verification control,
along with improved programming yield and reliability.
The 1T bit cell's reduced size results
in higher yield and lower product costs for OTP customers. Additionally,
since oxide defects are directly related to area, the smaller 1T cell
inherently provides greater quality and reliability than a 1.5T or 2T
cell.
Lower Power and Higher Performance
In addition to improved density and reliability,
the 1T-Fuse cell enables significant power reduction and speed improvement.
Sidense's Low Power (SLP) architecture targets RFID and other ultra low power applications such as medical implantable devices. Additionally, the sub-10ns Sidense SiPROM architecture is the fastest embedded NVM solution in the industry.
The Clear Choice
With the smallest cell area, fastest access time and the lowest power consumption, Sidense 1T OTP provides significantly better reliability and programmability than any alternative NVM solutions. Sidense 1T OTP is the clear choice for today's consumer and wireless markets.
Search for Sidense IP here
About the Author
With over 25 years of IC design and manufacturing experience, Wlodek Kurjanowicz has led numerous design, design automation and design analysis groups. In 2003 he founded Sidense Corp. with a vision to build a reliable NVM solution for sub 100nm CMOS technologies. In 1998 he co-founded ATMOS Corp., the embedded memory IP company, which he lead as its Chief Technology Officer to become the world leading provider of embedded DRAM compilers. Wlodek became a Mosys Fellow following the ATMOS acquisition by Mosys Inc. (Nasdaq: MOSY) in 2002. Prior to that, he managed the Design Analysis Group and held the position of Senior Technical Advisor at Chipworks, and was a Member of Technical Staff at Semiconductor Insights Inc. He also held various IC Design Manager and IC Technology Manager positions in a Polish semiconductor plant. He holds six patents in addition to several pending applications in the memory IP space.