next up previous contents
Next: 6. Gate Delay Time Up: 5. Peaking Channel Doping Previous: 5.6 Practical Considerations


5.7 Alternative PCD Structures

When looking at the PCD the question arises whether it is possible to extend the highly doped channel area to the source well without losing too much in performance. To investigate this an optimization is performed for Device Generation A which is very similar to the PCD optimization with one Gaussian function (Method 1 in Section 4.4.3), but this time the doping is extended to the left side into the source well.

This doping profile can be realized, for example, by a Large Angle Tilted Implantation (LATI) as shown in Fig. 5.15 where acceptors are implanted from the source side after the source and drain wells have been implanted [12,51]. This would either require an additional mask step with the source window opened only and a rotating wafer during ion implantation, or uniform orientations of all the transistors on the wafer with a fixed implantation direction using the same mask as for the source and drain wells.

Figure 5.15: The LATI method to generate the modified PCD structure.
\resizebox{0.5\textwidth}{!}{
\psfrag{G} [bc][bc] {Gate}
\psfrag{S} [bc][bc] {So...
...[bc][bc] {LATI}
\includegraphics[width=0.5\textwidth]{../figures/latid-pcd.eps}}

Fig. 5.17 shows the resulting acceptor doping profile of this modified PCD device. The electrical characteristics marginally differ from the PCD device, therefore the drive current improvements of the two structures are fairly the same.

To avoid the asymmetry of the PCD device the implantations could be provided from both sides, source and drain. The result from another optimization procedure using two symmetric Gaussian functions (Fig. 5.16) shows that the performance gain is only about half as high: The drive current is 316.5 $\mu $A and the performance gain with respect to the uniformly doped device is 22.4% compared to 42.9% of the asymmetric structure.

Figure 5.16: The optimization result of the symmetric modified PCD device for Device Generation A.
\resizebox{\textwidth}{!}{
\psfrag{x [um]} [ct][cb]{$x$\ ($\mu$m)}
\psfrag{y [um...
...
\includegraphics[height=\textwidth,angle=90]{../figures/top-symmetric-pcd.eps}}
\resizebox{0.95\textwidth }{!}{
\psfrag{x [um]} [ct][cb]{$x$\ ($\mu$m)}
\psfrag{...
...cludegraphics[height=0.95\textwidth ,angle=90]{../figures/3D-symmetric-pcd.eps}}

If the two Gaussian functions from both sides overlap, they cause a doping peak in the middle of the channel as shown in Fig. 5.18 as the result of a further optimization run. The benefits of an asymmetric PCD device structure do not apply in this case, but the improvement in drive current compared to the uniformly doped device is still

Figure 5.17: The optimization result of the modified PCD device for Device Generation A.
\resizebox{\textwidth}{!}{
\psfrag{x [um]} [ct][cb]{$x$\ ($\mu$m)}
\psfrag{y [um...
...te}
\includegraphics[height=\textwidth,angle=90]{../figures/top-finger-pcd.eps}}
\resizebox{0.95\textwidth }{!}{
\psfrag{x [um]} [ct][cb]{$x$\ ($\mu$m)}
\psfrag{...
...\includegraphics[height=0.95\textwidth ,angle=90]{../figures/3D-finger-pcd.eps}}

remarkable. The drive current is 334.5 $\mu $A and the performance gain with respect to the uniformly doped device is 29.4% compared to 42.9% for the standard asymmetric PCD device.

Figure 5.18: The optimization result of the overlapping PCD device for Device Generation A.
\resizebox{\textwidth}{!}{
\psfrag{x [um]} [ct][cb]{$x$\ ($\mu$m)}
\psfrag{y [um...
...te}
\includegraphics[height=\textwidth,angle=90]{../figures/top-middle-pcd.eps}}
\resizebox{0.95\textwidth }{!}{
\psfrag{x [um]} [ct][cb]{$x$\ ($\mu$m)}
\psfrag{...
...\includegraphics[height=0.95\textwidth ,angle=90]{../figures/3D-middle-pcd.eps}}

A disadvantage of this method using two overlapping Gaussian functions is that the ratio between the maximum and minimum doping values along the channel is always smaller than two which limits the performance improvements compared to the standard PCD device, even in the asymmetric case. For the standard PCD device the ratio between background doping and peak doping is almost two orders of magnitude (Table 4.3).

The substrate doping has been an optimization parameter for the modified PCD device optimization where the doping peak was extended into the source well. To simplify the PCD structure shown in Fig. 4.10 where the substrate doping is kept constant at 10$^{15}$ cm$^{-3}$ and a second Gaussian function was used to suppress punchthrough (Method 2), a source halo optimization is performed. One Gaussian function is used in this case which defines the channel doping and provides the doping layer under the source well to prevent the punchthrough at the same time.

Fig. 5.19 shows the resulting acceptor doping profile. The vertical peak length is quite large to reach under the source well. The source halo device has a worse performance than the corresponding PCD device. Its drive current is 336.1 $\mu $A and, therefore, by 10% smaller than the PCD device (Method 2). Anyway, it still offers a remarkable performance gain compared to the uniformly doped device or the device resulting from one-dimensional optimization. It has been successfully demonstrated in the literature that this device structure can improve the performance of low-power applications [10,35].

Figure 5.19: The optimization result of the source halo device for Device Generation A.
\resizebox{\textwidth}{!}{
\psfrag{x [um]} [ct][cb]{$x$\ ($\mu$m)}
\psfrag{y [um...
...Gate}
\includegraphics[height=\textwidth,angle=90]{../figures/top-halo-pcd.eps}}
\resizebox{0.95\textwidth }{!}{
\psfrag{x [um]} [ct][cb]{$x$\ ($\mu$m)}
\psfrag{...
...
\includegraphics[height=0.95\textwidth ,angle=90]{../figures/3D-halo-pcd.eps}}

Table 5.5 lists the optimized parameters of the modified PCD device, the symmetric modified PCD device, the overlapping PCD device, and the device with a source halo. Their performance is compared to the uniformly doped device and the standard PCD in Table 5.6.


Table 5.5: Optimized parameters of the various practical alternatives of the PCD device
parameter unit mod. PCD symm. mod. PCD overl. PCD source halo
$N_{\mathrm{sub}}$ cm$^{-3}$ 2.62$\cdot$10$^{16}$ 1.24$\cdot$10$^{16}$ 2.16$\cdot$10$^{16}$  
$N$ cm$^{-3}$ 1.60$\cdot$10$^{18}$ 1.04$\cdot$10$^{18}$ 6.98$\cdot$10$^{17}$ 1.25$\cdot$10$^{18}$
$y_0$ $\mu $m 0.01898 0.01830 0.01724 0.02100
$\Delta x$ $\mu $m 0.25347 0.24671 0.30000 0.22824
$\sigma_x$ $\mu $m 0.01000 0.01737 0.01270 0.01000
$\sigma_y$ $\mu $m 0.01000 0.01676 0.0100 0.03272


Table 5.6: Performance comparison of the various practical alternatives of the PCD device
device $I_{\mathrm{on}}$ ($\mu $A) perf. gain
uniformly doped 258.5 -
standard PCD 369.3 42.9%
modified PCD 367.3 42.1%
symmetric modified PCD 316.5 22.4%
overlapping PCD 334.5 29.4%
source halo doping 336.1 30.0%

This comparison shows that the basic idea of the PCD can successfully be used to find alternative structures which are closer to manufacturability than the conventional PCD device. Their drive performance gains are still remarkable especially for the asymmetric doping structures that have been presented.


next up previous contents
Next: 6. Gate Delay Time Up: 5. Peaking Channel Doping Previous: 5.6 Practical Considerations
Michael Stockinger
2000-01-05