Fermi-FET transistor technology can lead to significant improvement in circuit performance, layout density, power requirements, and manufacturing cost with only a moderate alteration of traditional MOSFET manufacturing technology. This technology makes use of a subtle optimization of traditional buried channel technology to overcome the known shortcomings of buried channel while maintaining large improvements in channel mobility. This technology merges the mobility and low drain current leakage of BCA devices as well as the higher short channel effect immunity of SCI devices. This paper highlights aspects of the technology in a non-mathematical presentation to give a sound general understanding of why the technology is the most promising avenue for advanced very short devices.
Transistor scaling, a major driving force in the industry for decades, has been responsible for the dramatic increase in circuit complexity. Shorter gate lengths have required lower drain voltages and concurrently lower threshold voltages. Recent CMOS evolution has seen a dramatic reduction in operating voltage as transistor size is reduced. This was due to the maximum field limit on the gate oxide needed to maintain good long-term reliability. Proper selection of the gate material can produce low threshold transistors with off-state performance parameters equivalent to high threshold devices.
The Buried Channel Accumulation device, currently being used for p-type transistor processes has the Fermi level at a considerable depth from the gate thereby making it difficult to shut the device off. Attempts to bring the Fermi level up result in severe degradation of device performance. Need for optimization of existing BCA technology arose and Thunderbird Technologies, Inc. delivered! The Ëœincredibleâ„¢: Fermi-FET.
The Fermi-FET technology brings the Fermi level nearer to the gate. This technology merges the mobility and low drain current leakage of BCA devices as well as the higher short channel effect immunity of SCI devices. This paper highlights aspects of the technology in a non-mathematical presentation to give a sound general understanding of why the technology is the most promising avenue for advanced very short devices.
Fermi-FET technology can lead to significant improvement in circuit performance, layout density, power requirements, and manufacturing cost with only a moderate alteration of traditional MOSFET manufacturing technology. This technology makes use of a subtle optimization of traditional buried channel technology to overcome the known shortcomings of buried channel while maintaining large improvements in channel mobility.
Fermi-FET can optimize both the N-Channel and P-Channel devices with a single gate material, provided the work function is near the mid-range between N and P-type polysilicon. Materials that have been used in MOSFET technology with a suitable work function include Tungsten, Tungsten Silicide, Nickel, Cobalt, Cobalt Silicide, P-type Gei and many others. There is about a 30% reduction in junction capacitance relative to traditional MOSFET devices. This fact alone gives a significant speed advantage to the Fermi-FET in large scale circuits. The total speed improvement produced by both the lowered threshold and lowered gate and junction capacitances is very substantial.
In order to illustrate the impact of lowered threshold voltages via work function engineering, the large-signal transient response of two inverter structures was simulated. A comparison of conventional CMOS and metal-gate Fermi-FET structures was performed. It is seen that the Fermi-FET inverter displays significantly improved rise and fall times compared to the MOSFET. The different delay characteristics are evident. It is seen that the Fermi-FET inverter displays significantly improved rise and fall times compared to the MOSFET
The individual device DC characteristics were already well-known from the device simulations. For each inverter, the supply voltage was ramped up to Vd with a delay sufficient to allow the circuit nodes to settle to their initial DC state with the input low. The input was then pulsed high, then low; again with a delay time long enough to guarantee all nodes reach steady state. The corresponding outputs obtained give a comprehensive view of the device performance as compared to the traditional technology and thus acts a primary assessment of the feasibility of the new technology in lieu of existing ones.
The output of the mixed-mode simulations is shown in the figure. Even at 0.4 mm gate length the low threshold Fermi-FET is almost twice as fast as the MOSFET in this simple circuit.
Simple circuits such as this underestimate the benefit of the lowered capacitance associated with the source/drain junctions, but they virtually ignore the capacitance associated with the extended wiring in large circuits. The Fermi-FET is the emerging technology in the ever-expanding empire of electronics circuits and devices and is slated to be crowned the king in foreseeable future.
THE TRANSISTOR STRUCTURE
The Fermi-FET is a unique patented variation of the broad class of devices known as Field Effect Transistors (FET). Although the transistor operation differs markedly from standard MOSFET devices, the structure of the new device has many similarities, thus permitting easy conversion of existing CMOS process lines to production of Fermi-FET transistors.
The basic principle behind the working of a Field Effect Transistor is the conducting semi-conductor channel between two ohmic contacts; source and drain. The gate terminal controls the channel current and is a very high-impedance terminal. The FET is thus a three terminal, unipolar device. The name Ëœfield effectâ„¢ is due to the fact that the current flow is controlled by potential set up in the device by an external applied voltage. There are two types of FETs â€œ JFET and MOSFET. The FET of interest here is the MOSFET.
The N-channel MOSFET has two lightly heavily doped n- regions diffused into a lightly doped p-type substrate; separated by 25 Ã‚Âµm.These n-regions act as source and drain. An insulating layer is grown over the surface. Metal contacts are made for the source and drain. A conducting layer of metal will act as the gate, overlaying the insulating layer over the entire channel region. Due to the presence of the insulating layer, the device is called Insulated Gate FET (IGFET) or Metal Oxide Semiconductor FET ( MOSFET).
Modern Complementary MOS (CMOS) processes incorporate polysilicon gate structures less than 0.25 micron long, with the most common process being 0.15Ã‚Âµm. At this geometry, and the standard 1.8 volt Vdd, oxide spacers and drain extensions are common. Most processes also make use of the oxide spacer to form salicide on the gate and diffusions to reduce the sheet resistance and to control the polytime constant on wide transistors.
SURFACE CHANNEL INVERSION DEVICES
Most short channel CMOS processes create SCI type transistors for both P and N-Channel devices. This decision has evolved as line widths attained shorter dimensions primarily due to the reduced short channel effect sensitivity of the SCI devices over the BCA transistor, traditionally used for the PMOS. Its because of the widely known control problems with deep buried channel transistor (BCA) technology that most short channel processes incorporate both n-type and p-type polysilicon gates to create surface channel inversion (SCI) devices for both transistor polarities.
Figure 1 â€œ A cross-section drawing of an N-Channel SCI MOSFET transistor. The polysilicon gate would be degeneratively doped n-type. The drain extension region is typically utilized to reduce short channel effects (SCE).
Figure 1 depicts the main features of a short channel NMOS SCI device. Certain other refinements aimed at reducing short channel effects such as pocket implants, SSR well, graded channel, and elevated diffusions have been ignored for simplicity.
The operation of surface channel transistors is relatively simple to envision. Figure 2 shows an enlargement of the channel region with the source end drain extension from Figure 1. As the gate electrode is biased from zero (fully off) toward the threshold voltage, the initial charge on the gate drives free holes in the substrate away from the gate region.
Figure 2 â€œ A close-up of the channel region in Fig. 1 near the drain side of the gate. The gate electrode has a positive bias but is below threshold voltage. The silicon depleted of mobile carriers is shown as a white area. The arrows represent the vertical field direction.
The combination of depletion charge on the gate and induced space charged depletion region in the substrate create a vertical electric field through the dielectric and into the substrate. This impedes carrier movement and hence decreases carrier mobility. This is a major disadvantage of SCI transistors affecting device performance to a large extent.
Device scaling has forced the gate oxide to become ever thinner and the average doping in the well region to increase. These lead to reductions in carrier mobility; the ease with which charges can move through the transistor producing drive current.
Figure 3 â€œ The same region as in Fig. 2, but the gate bias is now just above the threshold. The depletion region has reached its maximum depth and a thin region of inverted silicon (n-type carriers in p-type silicon) exists just below the gate oxide. The arrows represent the vertical field and consist of two parts. The first part supports the depletion region and the second part reflects the carriers in the inversion layer.
As the gate bias continues more positive, the threshold voltage is reached. At this point the depleted silicon region under the gate stops expanding and additional charge on the gate results in free conduction carriers from the diffusion moving to the region under the gate (see Figure 3). It is these carriers that are responsible for the output current of the transistor.
PROBLEMS WITH SCI
It should be noted that for an SCI device, good short channel performance requires high channel doping to counteract short channel induced leakage and threshold drop, but at the same time needs a low threshold voltage due to the low operating voltages of short channel processes. This dictates the thinnest possible gate oxide must be used.
The thin gate oxide and high vertical electric fields have caused a new challenge not previously large enough to cause difficulties; polysilicon depletion. Polysilicon depletion occurs when free carriers are swept away from the bottom of the poly gate due to high vertical fields. In an SCI type of device, this occurs when the transistor is fully turned on. As can be seen in Figure 3, this depletion causes the gate dielectric to appear thicker than it actually is, reducing transistor performance.
In summary, SCI devices, the current device used in many short channel applications, have three major design difficulties:
Â¢ High Capacitance â€œ As SCI devices shrink, the higher substrate doping causes increased parasitic junction capacitance around the source and drain regions.
Â¢ Lowered Channel Mobility â€œ Higher channel doping and increased vertical electric field make it more difficult for carriers to move across the channel.
Â¢ Polysilicon Depletion â€œ The high charge on the gate can cause polysilicon depletion to occur, lowering the transistor drive current.
BURIED CHANNEL ACCUMULATION DEVICES
At longer channel widths, the BCA architecture was widely used for the P-Channel transistor in CMOS processes. This was primarily done because the BCA transistor would use the same n-type polysilicon gate used by the N-Channel device, greatly simplifying the process. Recently it became apparent to most manufacturers that the BCA architecture was incapable of scaling to the very fine line widths in development today. The added process complexity of using both n-type and p-type poly was offset by the better SCE immunity of the SCI transistor.
Figure 4 â€œ A cross-section drawing of an N-Channel BCA transistor. Note that the polysilicon gate would be doped p-type. The drain and source extensions are connected by an n-type channel layer. An N-Channel BCA device would be built with the structure shown in figure 4. The polysilicon electrode is p+, and there is an n-type channel between the source and drain. The depth of the channel is minimized to reduce short channel effects (SCE).
Although the structure of a BCA device is very similar to a SCI transistor, the operation of the two is markedly different. This is due to the presence of the p-n junction formed by the channel and well regions abutting. As in all p-n junctions, free carriers diffuse across the junction until the retarding field, due to the ionized donor and acceptor atoms, causes the drift and diffusion flows to equalize. This field leads to the built-in potential of p-n junctions. The magnitude of this built-in potential is determined by the doping density of the silicon on both sides of the junction.
This potential causes carrier depletion on either side of the metallurgical junction. The widths of these depletion regions are proportional to the relative doping of the p and n regions. In traditional BCA architectures, the channel region is more highly doped than the well region beneath it. The arrows in Figure 6 show the vertical electric fields present with the gate electrode at zero bias.
Figure 5 â€œ A close-up of the channel region from Fig. 4 above.
The dashed line depicts the location of the p-n junction and depleted silicon is shown as a white area. The arrows represent the vertical field direction. Note that the junction potential is not high enough to fully deplete the entire channel region. There is another field due to the gate work function that also depletes the surface part of the channel region.
This potential also causes some depletion of charge near the bottom surface of the gate. Since the gate doping is usually very high the depletion width in the gate is small, but as channel lengths shrink, this can become an important effect in the subthreshold region.
Figure 6 â€œ A graph depicting the vertical electric field on a line normal to the silicon surface under the conditions of Figure 5. Wjn and W jp make up the space-charge region due to the metallurgical junction. The region labeled Wpn is depleted due to the potential difference between the gate and the channel. This charge is reflected in the poly depletion Wpp .
Figure 6 shows the vertical electric field in the wafer with the gate electrode at zero bias (as in Fig. 5). Note that the field reverses direction within the channel region where the depletion from the junction Wjn meets the depletion induced from the gate electrode Wpn.
As the gate electrode bias is moved from zero towards Vdd , the field from the gate initially decreases reaching zero at Vt . The gate field then reverses and begins to climb as more charge is injected into the channel at the source electrode. Figure 7 shows the conduction channel forming just above the junction depletion region.
Figure 7 â€œ The channel region from Fig. 5 above, with the gate bias at just above the threshold voltage. The channel begins to open near the edge of the junction depletion region and widens toward the surface as neutral silicon until the surface is reached. High-level injection begins after this point.
ADVANTAGES OF BCA
BCA devices have a significant advantage over SCI devices in terms of channel mobility. This is due to two reasons.
Â¢ First, the vertical field within the channel is substantially lower in a BCA device. This is because the gate does not have to deplete majority carriers away from the interface to form a channel. The gate supplied vertical field is due only to the mobile conduction charge.
Â¢ Second, improvement in channel mobility because this architecture can be made with a significantly lower total doping density in the channel. This improved channel mobility is a major advantage over traditional surface channel inversion devices.
Figure 8 â€œ The vertical electric field on a line normal to the silicon surface under the conditions of Figure 7. Note the neutral region above Wjn below Wpn . Increasing the gate bias will reduce Wpn and Wpp but leave the junction field largely unchanged.
Figure 8 shows the vertical electric field associated with a transistor in the condition shown in Figure 7. Further increases in gate bias result in the channel width increasing toward the surface and then excess carriers are injected with an exponential distribution peaked slightly beneath the interface, as shown in Figure 9.
Figure 9 â€œ With sufficient gate bias, the channel is under high-level injection. Since the carriers are majority type the only vertical field present within the conducting layer is that due to the conduction carriers themselves.
Note that the junction depletion region remains largely unchanged during the gate sweep. Figure 9 shows the vertical field profile with the gate bias near Vdd. Even though the conduction carrier distribution at this point has the same exponential shape as an SCI device, the distribution width is greater in the BCA and the peak concentration is lower. This decreases carrier-carrier interactions also helping BCA channel mobility.
Another benefit of this type of device is that the poly depletion occurs at the off state. Poly depletion negatively affects leakage but not drive current. This is in itself a rather important development and hence needs to be optimized further to suit growing demands on semi-conductor technology.
Figure 10 â€œ The vertical electric field under the conditions of Figure 9. Depending upon the lateral field conditions there may or not be any region above the junction with a zero vertical field component.
PROBLEMS WITH BCA
At longer channel widths, the BCA architecture is superior to surface channel architectures. The improved channel mobility leads to higher saturation current, the lower substrate doping can dramatically reduce the parasitic diffusion capacitance and the wider channel lowers the gate capacitance at equivalent oxide thicknesses. Problems with this type of transistor appear as the channel length and operating voltage decrease.
Transistor scaling, a major driving force in the industry for decades, has been responsible for the dramatic increase in circuit complexity. Shorter gate lengths have required lower drain voltages and concurrently lower threshold voltages. The problem with the BCA is that the channel turns on at the bottom of the conducting channel, far removed from the gate electrode as shown in Figure 7. With the initial conduction far from the surface it can be difficult to shutoff the device.
Figure 11 - Atlas output predicting the N-Channel Id Vg performance of identical BCA MOSFET transistors at two different L o values.
Notice in Figure 11 that the subthreshold region of the shorter device is no longer linear. The two devices show a Vt of a reasonable 0.08V, but as the gate bias approaches zero volts, there is a distinct non-linearity to the shorter line width subthreshold swing. The huge rise in leakage produced by this effect limits the usefulness of the device to devices above a minimum line width.
Transistor scaling requires the operating voltage of devices to be reduced as well. SCI devices are able to reduce threshold by thinning the gate oxide thickness. BCA devices have only a very small but negative correlation between Vt and Tox . Threshold can be reduced in a BCA by increasing the channel depth, which augments the problem above as the channel opens even further from the gate as the threshold is dropped. Threshold can also be reduced by increasing the channel doping, keeping the depth constant. This improves the 2D degradation, but the mobility reduction with the added doping removes the drive current advantage.
Figure 12 - Atlas output predicting the N-Channel Id Vg performance of BCA MOSFET transistors with reduced thresholds by increasing doping in the channel region.
The problems associated with BCA devices all spring from the fact that the channel opens from the bottom, significantly removed from the gate. This lack of gate coupling leads to the poor subthreshold swing and limits the minimum attainable channel length. These problems can b esummarised as
Â¢ Minimum attainable channel length is limited-as the depth of the channel is reduced, depletion regions attain comparable dimensions
Â¢ The BCA transistor is difficult to turn off- the opening is at the bottom of the channel, far removed from the gate electrode
Â¢ The sub-threshold swing is non-linear- the sub-threshold characteristics of BCA devices is poor
SURFACE CHANNEL ACCUMULATION
The channel of a BCA device first opens at the edge of the channel side depletion region caused by the p-n junction (see Figure 7). If the doping density in the channel layer is lowered, or the layer is made thinner, the channel will open closer to the gate oxide interface. Eventually there will not be enough dopant in the channel layer to completely satisfy the space-charge requirement of the junction below it. With less dopant in the channel than the junction requires, the conduction path will open at the silicon surface similar to a SCI device, but the conduction in this type of transistor will be majority rather than minority carriers. Hence the name Surface Channel Accumulation or SCA transistors.
Figure 13 â€œ Long channel simulations showing the Subthreshold Swing vs. Xj for three different Nd :Na combinations . As the junction approaches the surface, the transistor changes from BCA type to Fermi-FET and then to SCA type.
SCA and BCA transistors make up the broad family of buried channel transistors. The Fermi-FET transistor is a specific optimization of buried channel transistors. A Fermi-FET transistor occupies the region near the crossover from BCA to SCA devices. This allows a Fermi-FET to have the mobility and subthreshold swing advantages of a BCA device while maintaining the short channel effect immunity of a SCI or SCA transistor.
Figure 13 shows the S parameter or Subthreshold Swing of long channel transistors using several combinations of substrate and channel doping densities. Regardless of the doping levels used, there is a minimum S value where the device transitions between BCA and SCA operation.This is the Fermi-FET region. This minimum value corresponds to the point where the device threshold is roughly equal to the Fermi-potential plus the built-in potential of the junction. The S-chart serves as a means to tune the profiles for maximum Fermi-FET performance.
Figure 14 shows simulation results of N-Channel transistors with gate lengths of 0.2 Ã‚Âµm. Both devices have identical gate oxides and Source/Drain diffusions. The Fermi-FET device uses p-type polysilicon for the gate electrode while the SCI device uses the traditional n-type. Because of the low drain voltage (1.8V), the Fermi-FET was moved slightly toward the BCA end of the range to reduce the threshold somewhat.
Even with a much higher Vt , the Fermi-FET device outperforms a similar surface channel device.The linear graph clearly shows the much improved mobility of the Fermi-FET. Both devices have identical diffusion profiles and contact spacing. The log plot of Figure 14 shows that there is a small amount of DIBL penalty for this device, but even at the BCA end of the threshold spectrum it is still a reasonable amount. Higher operating voltages allow an even greater performance improvement due to increased overdrive, but lower operating voltages (further scaling) require threshold voltage lowering techniques.
Figure 14 â€œ Simulation of identical SCI (dashed line) and Fermi-FET (solid line) transistors produced these IdVg curves
Figure 15 depicts a Fermi-FET transistor analogous to the BCA device shown in Figure 7. The difference being that the doping in the channel region is lowered (or the depth is decreased) such that the depletion region from the p-n junction would extend to the region very near the surface of the silicon.
Figure 15 â€œ A close-up of the channel region of a Fermi-FET type of transistor. This drawing is similar to the BCA shown in Fig. 5. The dashed line depicts the location of the p-n junction and depleted silicon is shown as a white area.
With the junction depletion using most or all of the dopant in the n-channel, an interesting aspect of SCA devices occurs. Equation 2 defines a built-in potential associated with a gate electrode added above the oxide in Figure 13. Movement of charge between the polysilicon gate and the silicon must balance this potential, but there is not enough charge left within the channel region to deplete and satisfy the potential difference.
Figure 16 illustrates how this extra charge movement is accomplished in a SCA device. Once all mobile electrons are depleted from the channel, additional potential increase is created by moving holes from the substrate into the formally n-type region, creating a volume inversion or a psuedo p-type region.
Figure 16 â€œ A simulation depicting the location and density of mobile charged carriers in the substrate of a Fermi-FET transistor. Electrons and hole concentrations are shown according to the scale at left. Only concentrations above ni are shown so the white regions designate depleted silicon. The Gate and Source are at zero volt bias with the Drain electrode at 1.5 V (right side).
This volume inversion is unique to the SCA (and Fermi-FET) type of device. It is responsible for the lack of short channel roll-off usually seen in BCA type transistors. With the gate potential at zero, the drain field cannot reach the source diffusion and cause carrier injection. The drawing in Figure 16 was produced using Silvaco Internationalâ„¢s Atlas program. The simulation was continued with increased gate bias through threshold into saturation. As the gatebias is increased slightly, the excess holes are swept away from the interface. The Figure ËœAâ„¢ uses the identical carrier concentration shading histogram shown in Figure 16.
Even 100 millivolts on the gate is enough to deplete the silicon surface region. At this gate bias, the drain current begins to rise although the current level and carrier concentration are still very low. Further increases in Vg continues this process. At V g =0.3V, the concentration in the channel is above n i , but still less than the chemical donor concentration in the channel region. The channel layer still has a positive space charge. This is shown in Figure ËœBâ„¢.
Note that the current flow while at the surface near the source is not tightly bound to the interface by a strong vertical field as in a SCI device. At the drain end of the channel, the current flow actually moves significantly away from the interface due to the relatively high drain bias. The low vertical field is one of the main contributors to the significant increase in channel mobility of this type of transistor.
At this gate bias, the channel region contains many more free electrons than holes, but the electron concentrations are still far short of the ionized donor atom population. Charge neutrality has not yet been reached, so no carrier accumulation in the channel has occurred. We define threshold of SCA or Fermi-FET devices as the point at which the channel carrier concentration exceeds the net chemical dopant concentration; the onset of strong accumulation.
The Figure ËœCâ„¢ shows still higher gate bias with a dashed white line representing the electron concentration of 1017 cm-3 at the source diffusion. 1017 is roughly the net dopant concentration in the channel so the line roughly represents the boundary between strong and weak accumulations. The transistor has not yet reached Vt , although the channel charge and drain current are both rising rapidly. The high threshold is a potential drawback of this type of device. If it were not possible to overcome this limitation, it would be unusable for low voltage applications.
At an applied gate bias of 0.9 V. the 1017 carrier concentration line is now continuous across the channel region. This situation corresponds roughly to a traditional extrapolated V t and results in a drain current of approximately 10 â€œ6 A/Ã‚Âµm. If the transistor falls into the Fermi-FET category, the vertical component of the electric field through the gate dielectric is close to zero at a gate bias of Vt . The very low vertical field results in near-bulk carrier mobility in the channel. This can be seen in the high gm values (about 3X higher than SCI) near threshold. Figure 17 shows a Fermi-FET transistor in saturation along with a surface channel device of identical size for comparison.
Figure 17 â€œ Simulation results of Fermi-FET (left) and SCI (right) transistors depicting the location and density of mobile charged carriers in the substrate. Both transistors are in saturation with Vg =Vd =1.5V.
Note in Figure 17 that the carrier type in the polysilicon gate is opposite since an N-channel Fermi-FET uses p-type polysilicon as a gate. Another point of interest is the large difference in the depletion width between the two transistors. This is due to the lower substrate doping levels used in a Fermi-FET and results in lower parasitic junction capacitance.
Figure 18 shows the simulated drain current output of the two transistors illustrated in Figure 17. The Fermi-FET clearly shows the potential for superior performance with a subthreshold swing of 73.3 mV/decade compared to 91.5 mV/decade for the SCI device, but the higher Vt of the Fermi-FET does not permit enough overdrive to match the saturation current seen in the inversion device.
Figure 18 â€œ ATLAS simulation results of Fermi-FET and SCI transistors depicting the IdVg response in both log and linear plots. Both transistors have identical gate lengths and dielectric thicknesses.
Short channel lengths have been accommodated in SCI and traditional BCA devices by reducing the applied drain voltage and the threshold voltage accordingly. The Fermi-FET however operates near the point where the channel junction just depletes the channel region as shown in Figure 15. If the threshold voltage is dropped by increasing the channel doping or making the channel deeper, the device will move into the BCA mode of classical buried channel transistors with itâ„¢s poor short channel performance.
As long as Vd is large compared to the threshold voltage Vt , (Vd 2.5 Vt ) the increased lateral carrier velocity in the channel of the Fermi-FET produces a higher saturation current. Without a suitable low threshold Fermi-FET, only higher voltage applications would benefit from the improved transistor architecture. Fortunately a viable alternative is available to permit operation at any threshold desired.
FERMI-FET BUILT-IN POTENTIALS
Minimizing the vertical component of the electric field in the channel is responsible for the improved mobility performance of the Fermi-FET. This optimization does however limit the range of threshold due to the substrate doping. Changes in channel doping are possible, in fact they are desirable for particular line widths and applications, but these changes must be accompanied by changes in the Well doping and/or channel depth to maintain the Fermi-FET low field condition. This means that many different channel doping levels will all have approximately the same threshold voltage.
In Fermi-FET devices, threshold reduction must be implemented in a different manner The factors responsible for the threshold voltage in a buried channel transistor (including the Fermi-FET transistor) are defined in Equation 3 below. Equation 3 represents the terms behind the threshold voltage of a buried channel transistor. The threshold voltage (neglecting the Vt term which is the result of short channel effects) is made-up of four distinct voltage components. These are shown graphically in Figure 19.
Figure 19 - Athena simulated cross section with the component voltages from the Vt equation.
Â¢ V 1 results from the difference in contact potential between the aluminum metal and the p-type polysilicon gate, and the aluminum metal and the p-well region under the fermi-tub structure.
Â¢ V 2 quantifies the voltage induced across the depletion region below the fermi-tub-well junction.
Â¢ V 3 represents the voltage across the fermi-tub itself. This is made up of the depletion region above the fermi-tub-well junction and depletion between the junction induced depletion region and the silicon surface due to the gate field.
Â¢ V 4 quantifies the voltage developed across the gate oxide due to the field from the polysilicon gate terminating on charge in the region defined as V 3 .
The Fermi-FET is a special case and can allow the equation to be simplified. Since the Fermi- FET condition requires the depletion region to just deplete the channel region, V 2 and V 3 reduce to the potential across the junction itself.
With the tub fully depleted by the junction in a Fermi-FET, there will be virtually no vertical gate field at threshold, so V4 reduces to zero. This means that the threshold voltage of a Fermi- FET transistor using a polysilicon gate reduces to just V 1 +V 2 +V 3 .
Equation 5 shows that lowering the threshold voltage of a Fermi-FET is best accomplished by changing the first term. Large changes to the second term would move the device away from the Fermi-FET region. This is opposite to the traditional methods of threshold control.
V 1 is basically the difference in work function between the gate electrode material and the silicon substrate just below the depletion region. Figure 20 shows I d Vg simulations with an identical substrate but different gate materials. Choosing a gate with the proper work function allows threshold voltages well within the BCA range for a polysilicon gate device, but with none of the short channel penalties associated with deep buried channels.
Figure 20 â€œ Simulated IdVg show the effect of different work functions.
It is clear from Figure 20 that almost any desired threshold could be attained by variations possible to the V1 term. More importantly, if a material with a work function near the midpoint of the silicon band-gap were used, the same material would serve for both the P-Channel and N-Channel transistors. Such materials are Tungsten, Titanium Nitride, Tungsten Silicide, and several others not shown in the graphic.
Metal-gated Fermi-FET structures have yet to be verified experimentally, but initial experiments appear to affirm the academic work that postulates that the Fermi-FET architecture maintains significant advantages over SCI devices at least through gate lengths of 50 nm. In addition, the lowered vertical field in the Fermi-FET produces dramatic reductions in gate tunneling currents through very thin gate dielectric layers.
The Fermi-FET device developed and patented by Thunderbird Technologies Inc. can provide significant performance advantages relative to conventional CMOS technologies for a variety of products. The discussion below is a brief overview of several product areas that is predicted to benefit most from the Fermi-FET technology.
In general, the Fermi-FET will directly benefit performance-driven digital products, due to the higher drive current and lower capacitances inherent in the device architecture. The Fermi-FET has also been evaluated with respect to some of the other considerations that impact technology decisions, such as process sensitivity, leakage, power consumption, temperature and noise characteristics, scalability and reliability. In addition, the costs, in terms of dollars, human resources and time-to-market constraints must be carefully weighed when considering any new technology introduction. Considering these issues, the following product groups is expected to benefit in terms of market differentiation and competitive advantage.
DSP - Due to the increasing ASSP nature of the DSP market, in contrast to the MPU market, designs are introduced based upon specific market demands. This allows easier introduction of a new device or process technology into the product line. An existing CMOS line may be run in parallel with a new Fermi-FET line to support existing products, while existing products are redesigned, if feasible, and new products are being introduced. The technical benefits to be derived from the Fermi-FET technology for DSP products are expected as follows:
a) The Fermi-FET offers significantly higher performance due to lower intrinsic device capacitances and better drive current. In the highly competitive DSP market, this is particularly important.
b) Considering analog content, the Fermi-FET should offer lower noise than conventional CMOS, due to its buried-channel nature. This has not been verified experimentally yet, but it is reasonable to expect based upon the device structure, and its physics of operation.
c) Again concerning analog behavior, the Fermi-FET provides a significantly higher small-signal transconductance just above turn-on compared to a surface-channel device. Combined with the lower intrinsic capacitances, the higher gm means higher fT devices.
d) Matching properties should exceed those of conventional surface-channel devices, again due to the device's buried-channel nature, the lower dependence of Idsat on Vt, and the low dependence of Vt on oxide thickness.
e) Temperature characteristics of the Fermi-FET have been measured and typically exceed those of conventional surface-channel devices, in terms of current drive degradation with temperature.
MPU - For standard MPU products which are more performance-driven, rather than price- driven, the Fermi-FET must be introduced at an appropriate point in the technology roadmap. The MPU products require the highest performance possible, but are also very sensitive to time-to-market requirements and need design information which is timely and as accurate as possible. Technology scalability is also extremely important. Scaling the device to very deep-submicron linewidths, while retaining the performance advantages at longer linewidths is possible. Thunderbird is currently working on methods to extract Fermi-FET design information from measured data and/or simulations in a way that is compatible with existing commercial EDA tool sets. This is critically important. Product designers need to learn how to design with the Fermi-FET as quickly as possible, and the device characteristics need to be incorporated within the design methodology as smoothly and efficiently as possible.
MCU - For product applications such as these which may be embedded, are already established as catalog parts and are more price-driven, rather than performance-driven, the Fermi-FET can provide nearly a generational leap in performance with very little (if any) investment in retooling required. It has been Thunderbird's experience that it is often possible to reduce process complexity, due to the lack of the LDD in the device structure, for example, hence lower the manufacturing cost. Analog bolt-ons to MCU parts will also enjoy the same benefits as the DSP products.
Logic/ASIC/FPGA/Gate Array - General purpose logic will also benefit from the Fermi-FET's higher performance, of course, but the cost of porting such low margin products to a new technology should be carefully considered. The most likely candidates will be the ASIC, and perhaps gate array products. For these applications, the stream of designs may be simply switched over to the new technology once it is qualified. Gate array base levels would need to be redesigned, to take advantage of the expected performance increase. It would be reasonable to expect to be able to shrink die size based upon the device performance, for a given linewidth. This of course leads to lower manufacturing costs.
SRAM - As with the general-purpose digital applications, SRAM would clearly benefit from the reduced device capacitances and higher drive current. Due to the higher performance, it would be possible to shrink cell and I/O dimensions, and subsequently decrease the die size. As with ASICs and gate arrays, decreased die size means lower costs. This is in addition to the benefits derived from increased yield due to device stability with manufacturing variations.
Analog/Mixed-Signal - Telecom, data conversion and networking products would all benefit from the characteristics mentioned in the DSP bullet. To date, no Fermi-FET analog blocks have been fabricated by Thunderbird, but the expectations of lower noise, higher transconductance with lower device capacitance, hence higher fT are reasonable based upon current device physics knowledge and simulated characteristics.
Power MOS - This is an area Thunderbird plans to explore in the near future. Due to the buried nature of the channel, for longer channel length devices, the current density capability of the Fermi-FET is greater than surface-channel CMOS. Due to the higher mobility experienced by the carriers in the channel, the Rdson at a given geometry will be lower than conventional surface-channel devices as well. This allows either die shrinks or increased drive capability for applications requiring a very low output impedance, such as motors and actuators.
In summary, the products which are probably in the best position to benefit from the Fermi-FET technology could be ranked as follows:
DSP products (ASSP and standard parts)
MPU products (CPUs)
MCU products (embedded or standard parts)
Networking products (ATM chips)
ASICs/Gate Array products
Analog/Mixed Signal products (Conversion, industrial control)
Power/RF products (not known at this time)
Of course the list above is not exclusive, and there are many more areas which could benefit from a high performance device architecture such as the Fermi-FET.Significant manufacturing advantages may surface as well, but this is simply a reasonable expectation since not enough silicon has been run to compile a statistical database for any of the Fermi-FET variants made to date. The expectation is based upon the limited silicon to date, and an intuitive understanding of the device sensitivities. Other advantages may surface as well, particularly with respect to the path-breaking low-threshold technology.
The Fermi- FET is the latest in emerging revolutionary transistor technologies. Initial experiments appear to affirm the academic work that postulates that the Fermi-FET architecture maintains significant advantages over SCI devices at least through gate lengths of 50 nm. In addition, the lowered vertical field in the Fermi-FET produces dramatic reductions in gate tunneling currents through very thin gate dielectric layers. These features and the inherent advantages of the Fermi- FET over the existing traditional transistor technologies renders it the most promising of all evolving developments.
The performance advantages gained by using the Fermi-FET will provide unique marketing opportunities, through distinct product differentiation, which is important in increasing market share. In addition, decreased manufacturing costs, improved yields and die shrinks provide a rapid return on investment. Finally, as the Fermi-FET continues to be scaled, the technology will provide a sustainable competitive advantage.
1. B. L. Austin. Performance Analysis and Scaling Opportunities of Bulk CMOS Inversion and Accumulation Devices. PhD thesis, Georgia Institute of Technology, MAY 2002
2. M. W. Dennen. Fermi-FET technology
3. Ben G. Streetman. Solid State Devices
1. INTRODUCTION 01
2. THE TRANSISTOR STRUCTURE 04
3. SURFACE CHANNEL INVERSION DEVICES 06
PROBLEMS WITH SCI
4. BURIED CHANNEL ACCUMULATION DEVICES 10
ADVANTAGES OF BCA
PROBLEMS WITH BCA
5. SURFACE CHANNEL ACCUMULATION 19
6. THE FERMIâ€œFET 20
FERMI-FET BUILT-IN POTENTIALS
7. APPLICATIONS 32
8. FURTHER DEVELOPMENT 36
9. CONCLUSION 37
10. BIBLIOGRAPHY 38
I extend my sincere thanks to Prof. P.V.Abdul Hameed, Head of the Department, Electronics and Communication Engineering, for providing me his invaluable guidance for the Seminar.
I express my sincere gratitude to my Seminar Coordinator and Staff in Charge Mr. Manoj K, for his cooperation and guidance in the preparation and presentation of this seminars.
I also extend my sincere thanks to all the faculty members of Electronics and Communication Department for their support and encouragement.
Shahar Al Minnath