Patent Application
20150221723
The present invention relates to a pair of MOS (metal-oxide-semiconductor) transistors, and a manufacturing method thereof. A MOS transistor, or MOSFET (for metal-oxide-semiconductor-field-effect-transistor) is an insulated-gate field-effect transistor in which the insulating layer between each gate terminal and the channel is oxide material.
Integrated circuits require using different types of transistors: short channel transistors for logic operations, small surface transistor for memories (SRAM) and large channel surface for analog transistors.
In an integrated circuit, transistors of the same type must have a low dispersion of their technical features. Requirement specification on this dispersion depends on the function of the transistors in the integrated circuit.
Typically, with analogic function such as current mirror, two transistors are supposed to have same electrical parameters. They are then called a pair of matched transistors.
A generally analyzed parameter to check the low dispersion between transistors of a pair is the threshold voltage difference between the two transistors of said pair, since variations between these threshold voltages cause efficiency fluctuations, increase electrical consumption, eventually decrease manufacturing efficiency and eventually threaten the circuit functionality.
For analog transistors, threshold voltage difference between matched transistors is typically desired to be less than 1 mV.
With bulk silicon technology, dispersion predominantly results from Random Dopant Fluctuation (RDF), which results from variation in the impurities (or dopants) concentration in the channel region of a MOSFET and can alter the properties of the transistor, especially the threshold voltage VT.
Due to its randomness, this dispersion source tends to decrease as the transistor surface increases, due to averaging of dopant concentration.
Thus, when a person skilled in the art wants to have matched transistors, i.e. with substantially similar features, he uses transistors with large surfaces, i.e. whose channel surfaces are large. In particular, large surface transistors are used in pair for structures defining current references, current mirrors, or current comparator.
With the fully depleted silicon on insulator (FD-SOI) technology, transistors are built upon a thin layer of silicon over a Buried Oxide (commonly called BOx), said thin layer of silicon being undoped, or fully-depleted. Consequently, the predominance of Random Dopant Fluctuation as a dispersion source is suppressed.
A new source of dispersion is however introduced with respect to bulk silicon: channel silicon thickness TSi of transistor channel may substantially vary among transistors. Since threshold voltage VT depends on channel silicon thickness TSi, variation of channel silicon thickness TSi causes variation of threshold voltage VT.
Accordingly, the height of the surface 13 represents the average channel silicon thickness under the surface S13 of a first transistor built on top of silicon layer 1, and arranged on the left-hand side of the pair in
For the sake of simplicity, depicted surfaces such as surfaces 13 and 14 will be assimilated to their respective transistors as a graphical way to represent the relation between channel surface and average channel silicon thickness for the transistors.
As seen on
For smaller transistors, i.e. with a small surface channel, threshold voltage VT variations induced by channel silicon thickness differences are not the predominant sources of dispersion. Indeed, random sources of dispersion such as Random Dopant Fluctuation (RDF) have a greater effect on the electrical parameters of the transistors.
This side effect is however especially detrimental for large surface transistors, i.e. with a large surface channel. Because they cover a large surface, the average channel silicon thicknesses of two transistors can significantly differ.
In
It shall be noted that this is the same with any technology wherein the substrate has a buried interface which delimits the active region thickness of the channel of transistors. The buried interface is a change of material which confines the charge carriers contributing to the conduction (e.g. electrons for N-MOS transistors or holes for P-MOS transistors) in the active region. Delimiting the active region depth of the channel means that the active region may not extend beyond said interface, due to differences in electrical characteristics between the material under said buried interface and the substrate.
For example, the buried interface may be an interface between the silicon of the channel and an insulating substrate or an interface between the silicon of the channel and a buried layer distinct from the substrate. Said buried interface also defines the channel silicon thickness of the transistor. Such thicknesses consideration will be referred to as channel silicon thickness in the following description. Preferably, the active region has a thickness under 15 nm.
Preferably, the active region is fully depleted in the depletion region, i.e. with a doping concentration substantially equals to or inferior to 1017 cm−3 for the depletion region, for example 3 nm under the gate oxide, and under 1018 cm−3 for the whole active region.
The buried interface can thus be defined by an insulating layer such as in the Silicon-on-Insulator technology, and the active region is then the thin film of silicon above the insulating layer. An exemplary embodiment uses the FD-SOI technology, and the buried interface is then the interface between the Box and the thin silicon film.
Besides FD-SOI, it can be a bulk silicon substrate wherein a silicon film constituting the channel is isolated by a strongly doped silicon layer, e.g. dopant concentration beyond 1018 cm−3 or 1019 cm−3.
The following description will be made with reference to FD-SOI transistors as a non-limitative example.
An object of the present invention is providing a pair of transistors wherein the electric characteristics of the transistors are matched, and the manufacturing method for producing them.
To achieve this object, the invention proposes according to a first aspect a pair of transistors on a substrate having a buried interface which delimits the active region thickness of the channel of said transistors, said transistors having each:
wherein each sub-transistor of a transistor of said transistor pair is spaced apart from at least one adjoining sub-transistor of the other transistor of said transistor pair by a distance D less than half the transistor channel width WG, said distance D between two sub-transistors being measured between the respective center of the channels of said sub-transistors, said sub-transistor of a transistor and said adjoining sub-transistor of the other transistor being aligned in the direction of their widths;
Exemplary embodiments of a pair of transistor according to the invention will be described hereafter with reference to the appended figures; the description and the figures also showing additional features which are also part of the invention, among which:
The invention also concerns an integrated circuit having at least one pair of transistors according to the invention.
According to another aspect of the invention, a method is provided for manufacturing pairs of transistors according to the first aspect, i.e. a pair of transistors on a substrate having a buried interface which delimits the active region thickness of the channel of said transistors, said transistor pairs having:
a) defining a desired overall threshold voltage variability σΔVT,
b) determining values of random threshold voltage variability σΔVTR and channel silicon thickness induced threshold voltage variability σΔVTsi permitting said desired overall threshold voltage variability σΔVT to be achieved,
c) determining values of overall transistor channel surfaces and sub-transistor channel surfaces permitting to obtain the values determined at step b),
d) determining the numbers of sub-transistors required to achieve the values determined at step c),
e) selecting a number of sub-transistors close to the minimum of required sub-transistors, thereby determining the corresponding value of the total transistor surface.
Other aspects, objects and advantages of the invention will become better apparent upon reading the following description of embodiments of the invention as non-limiting examples, made with reference to the appended drawings wherein:
Channel silicon thickness variation is not entirely random. Respective thicknesses of two remote points are likely to be more different than respective thicknesses of two neighboring points. In other words, the more spaced apart the devices, the higher the silicon channel thickness statistical difference between them is. This can be seen with the large surface transistors of
The invention provides a pair of transistors on a substrate having a buried interface which delimits the active region thickness of the channel of said transistors, said transistors having each:
According to the invention, each transistor of a transistor pair is made of several interconnected sub-transistors. In other words, all the gates of the sub-transistors of a transistor are electrically connected together, all the sources of the sub-transistors of a transistor are electrically connected together, and all the drains of the sub-transistors of a transistor are electrically connected together. Accordingly, the interconnected sub-transistors thus behave like the corresponding large transistor.
These sub-transistors are arranged so that the average channel silicon thickness difference ΔTSi between the two transistors is minimized. The threshold voltages of the two transistors of said pair thus become substantially equal because of the minimization of the dispersion due to channel silicon thickness TSi. Further, by keeping a large overall channel surface, the random sources of dispersion are still averaged.
The transistors 33, 34 have the same transistor channel length LG, said transistor channel length LG being greater than or equal to 0.25 μm, the same transistor channel width WG, said transistor channel width WG being greater than or equal to 0.5 μm, and the same channel surface S, said transistor channel surface S being greater than or equal to 0.5 μm2.
Each of the transistors 33 and 34 are made of several sub-transistors, namely made of sub-transistors 43 for the transistor 33 and made of sub-transistors 44 for the transistor 34.
Each sub-transistor of a transistor has a sub-transistor channel length Li and has a sub-transistor channel width Wi. Said sub-transistor channel length Li is substantially equal to the transistor channel length LG of said transistor.
Said sub-transistor channel width Wi is smaller than the transistor channel width WG of said transistor, so that the sum of the sub-transistor channel widths Wi of the sub-transistors of a transistor is substantially equal to the channel width WG of said transistor. Accordingly, if N is the number of sub-transistors:
The sub-transistor channel width Wi is also smaller than the transistor channel length Lg. As a consequence, since the sub-transistor channel length Li is substantially equal to the transistor channel length LG of said transistor, the sub-transistor channel width Wi is also smaller than the sub-transistor channel length Li.
The configuration shown on
Since each sub-transistor of a transistor of said transistor pair is near at least one adjoining sub-transistor of the other transistor of said transistor pair, the channel silicon thickness difference between said sub-transistors is statistically minimized since the thickness variation between close points is more likely to be smaller than for points which are remote from each other.
For each sub-transistor 43, 44, there is a sub-transistor of the other transistor which is spatially close and have a similar channel silicon thickness. Since the channel silicon thickness difference ΔTSi between the two transistors is the sum of the thickness difference between their respective sub-transistors, the channel silicon thickness difference ΔTSi between the two transistors is also statistically minimized. Accordingly, the threshold voltage difference due to the channel silicon thickness difference is also statistically minimized.
Each sub-transistor 44 of the second transistors of the pair comprises a source S2, a drain D2, and a gate G2. The sources S2 of the sub-transistors 44 of the second transistor of the pair are connected together. The drains D2 of the sub-transistors 44 of the second transistor of the pair are connected together. The gates G2 of the sub-transistors 44 of the second transistor of the pair are connected together, although such connection is not shown in
The sub-transistors 43, 44 are aligned in the direction of their channel width. Since the channel width Wi of the sub-transistors 43, 44 is shorter than their channel length Li, aligning them in the direction of their channel width allows the channel center of two adjoining sub-transistors to be closer. Preferably however, in order to keep a good electrical insulation between them, the channels are spaced apart by a distance of a least 30˜50 nm.
The sub-transistors 43 of the first transistor and the sub-transistors 44 of the second transistor are interleaved so that each sub-transistor of a transistor of the transistor pair is adjacent to a sub-transistor of the other transistor of the pair.
The sub-transistors 43, 44 are arranged so that the channel of a sub-transistor of a transistor of the pair is adjacent to the channel of a sub-transistor of the other transistors, without any source or drain between them.
In
The most important point is the short distance between two sub-transistors of different transistors. Each sub-transistor of a transistor of the transistor pair is spaced apart from at least one adjoining sub-transistor of the other transistor of the transistor pair by a distance D less than half the transistor channel width WG, said distance D between two sub-transistors being measured between the respective center of the channels of said sub-transistors.
Distance D should be as small as possible. At least, D is less than half the transistor channel width WG, but preferably D≦WG/4, more preferably D≦WG/10.
For example, D≦1 μm, more preferably D≦0.1 μm. Smaller distances can be chosen, for example, for 14 nm node technology, a distance D of 64 nm, corresponding to the contacted pitch of this technology node.
With a short distance D, channel silicon thickness variation affecting a sub-transistor of a transistor affects similarly at least one sub-transistor of the other transistor. Therefore both transistors of the pair are affected similarly by channel silicon thickness variations.
This distance D should be as small as possible. Therefore, any sub-transistors of a transistor and the adjoining sub-transistor of the other transistor are preferably aligned in the direction of their widths, as shown in
There is no need for all the sub-transistors to be aligned along the same line. As shown in
Preferably, the sub-transistors are formed in groups with the same number of sub-transistors of each transistor. As shown in
Dividing the transistors 33, 34 in sub-transistors thus allows for a better matching of their characteristics. Preferably, each transistor is made of at least 3 sub-transistors, preferably at least 10 sub-transistors, more preferably at least 100 sub-transistors.
The transistors of the pair are not necessarily made of the same number of sub-transistors, but having the same number of sub-transistors facilitates their arrangement design and provides better result with respect to channel silicon thickness difference minimization. Accordingly, the two transistors of the transistor pair are preferably made of the same number of sub-transistors
Sub-transistors can have different widths Wi but having the same width Wi for sub-transistors facilitates their arrangement design and provides better result with respect to channel silicon thickness difference minimization. Accordingly, the sub-transistors preferably have the same width Wi.
According to the invention, lower threshold voltage variability can be obtained. Threshold voltage variability in a transistor is the statistical sum of the random and non-random components of this threshold voltage variability. Random components comprise for example the Random Dopant Fluctuation (RDF), which results from variation in the impurity (or dopant) concentration in the channel region of a MOSFET, the Line Edge Roughness fluctuation, affecting channel width and length and the fluctuations of work function in the gate stack.
Random components affecting the threshold voltage will be denoted σVTR, with the random variability between two transistors being denoted σΔVTR. Due to their randomness, these random components are averaged with the use of large transistors. The random part of variability is thus reduced with larger transistors. According to the invention, transistors thus have a surface larger than 0.5 μm2, preferably 2 μm2, and more preferably 5 μm2 or 10 μm2.
With an FD-SOI transistor, the channel is lightly doped, i.e. less than about 1017 cm−3, in the inversion zone, typically within 3 nm under the gate oxide. Accordingly, random dopant fluctuations are negligible. However, as explained above, channel silicon thickness fluctuation can alter transistor matching. Let threshold voltage variability due to channel silicon thickness be denoted σΔVTSi, and random variability between two paired transistors be denoted σΔVTR.
Overall variability of the threshold voltages between two transistors is then:
σΔV
with σΔVTSi being a function of the channel silicon thickness variability σTSi and of the sensitivity of the threshold voltage VT to channel silicon thickness variation (dVT/dTSi).
Channel silicon thickness fluctuation as a function of the distance X between two locations can be determined from thickness measurement data. Fourier transform of the channel silicon thickness measurements with respect to the location provides the power spectral density (PSD). Integration of this power spectral density on the spatial frequency-domain [X−1; ∞[ provides the average thickness variation σTSi.
As a transistor pair is considered, X can be approximately interpreted as the size of a transistor, possibly completed with the distance between two transistors.
As a non-limiting example, the graph of
As can be seen on
According to the invention, each transistor of the transistor pair, having a large channel surface, a channel length LG and a channel width WG, is divided into n smaller transistors, i.e. sub-transistors, preferably having the same length L and having a width equals to W/n. The sub-transistors of a transistor are then interleaved or mixed with sub-transistors of the other transistor of said pair.
The terminals of the sub-transistors of a transistor are interconnected. In other words, all the gates of the sub-transistors of a transistor are electrically connected together, all the sources of the sub-transistors of a transistor are electrically connected together, and all the drains of the sub-transistors of a transistor are electrically connected together. Accordingly, said interconnected sub-transistors thus behave like the corresponding large transistor. Thus, the larger transistor pair is replaced with n pairs of smaller sub-transistors.
These pairs of sub-transistors can be distributed on the substrate, but the distance between two sub-transistors of the same pair must be as small as possible, and preferably must tend to W/n.
Hence, the overall surface of the large transistors of the pair is still W×L, and the effect of random components is kept low. Furthermore, the distance between the large transistors of the pair corresponds to the mean distance between adjoining sub-transistors. Accordingly, the mean distance between the large transistors of the pair tends to be W/n. As a consequence, the contribution of channel silicon thickness variation to overall threshold voltage variability is reduced.
For instance, each transistor of a pair has a channel surface of 1 μm2, with LG=WG=1 μm. According to the example illustrated on
By providing a pair of transistors wherein each transistor is constituted by 10 sub-transistors, each of said sub-transistors having a channel length Li of 1 μm and a channel width Wi of 1.0 μm, then the overall channel surface of each transistor is still 1 μm2, corresponding to a random variability σΔVTR of 1 mV, but the mean distance between the two transistors of the pair is lowered down to 0.1 μm, corresponding to a channel silicon thickness induced variability σΔVTSi of 0.2 mV.
According to equation 1, the overall variability is thus reduced to 1.2 mV:
√{square root over (0.22+12)}≈1.2
As shown by this example, a large number of sub-transistors allows for a significant reduction of the threshold voltage variability. However, realizing a great number of sub-transistors can be difficult to achieve, whereas a low threshold variability may not be required.
The minimum number and size of the sub-transistors can be determined by the following method to achieve a desired overall variability, while having the fewer possible sub-transistors:
a) defining a desired overall threshold voltage variability σΔVT,
b) determining values of random threshold voltage variability σΔVTR and channel silicon thickness induced threshold voltage variability σΔVTsi permitting said desired overall threshold voltage variability σΔVT to be achieved,
c) determining values of overall transistor channel surfaces and sub-transistor channel surfaces permitting to obtain the values determined at step b),
d) determining the numbers of sub-transistors required to achieve the values determined at step c),
e) selecting a number of sub-transistors close to the minimum of required sub-transistors, thereby determining the corresponding value of the total transistor surface.
Preferably, the channel surface of a transistor and the number of sub-transistors are determined to achieve an overall threshold voltage variability σΔVT corresponding to
σΔVT=√{square root over (σΔVTR2+σΔVTSi2)}
As another non-limiting example for illustrating the method, in a similar way to
The pairs of random threshold voltage variability and channel silicon thickness induced threshold voltage variability, i.e. (σΔVtr; σVTSi) which allow for the desired overall threshold voltage variability σΔVT to be obtained is first determined.
As an example,
In this example, the dotted line 106, representing the required number of sub-transistors, shows a minimum for an overall surface of around 0.25 μm2. According to the dashed line 105, an overall surface of 0.25 μm2 corresponds to sub-transistor surfaces of about 600 nm2. As a result, around 417 sub-transistors of 600 nm2 are required for each transistor in order to achieve an overall threshold variability of 3 mV.
The effect of the invention can even be improved by keeping the channel silicon thickness induced variability σVTSi at a low level for sub-transistor surfaces.
With a prior art manufacturing method, the channel silicon thickness induced variability σVTSi must be kept low for the large surface of the transistor (typically above 0.5 μm2), which is difficult.
According to the invention, the channel silicon thickness induced variability σVTSi must only be kept relatively low for sub-transistor surfaces which are much smaller than the large transistor surfaces and thus much easier to achieve.
| Number | Date | Country | Kind |
|---|---|---|---|
| 1257792 | Aug 2012 | FR | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/EP2013/066930 | 8/13/2013 | WO | 00 |
