Professional Documents
Culture Documents
1-s2.0-S0026269223001969-main
1-s2.0-S0026269223001969-main
Microelectronics Journal
journal homepage: www.elsevier.com/locate/mejo
Index terms: This work aims to discuss the challenges of implementing an integrated ultra low voltage start-up clock/oscillator,
Energy harvesting the state of the art and propose four new variants of a body-biased stacked inverter based ring oscillator and
Ultra-low voltage analyse the same. The proposed delay cells designed in 180-nm BCD CMOS process are connected to form a
Ring oscillator
13-staged ring oscillator (RO) with regular 𝑉𝑡 transistors.
Body-bias
All four proposed variants’ performance is compared against the former works (implemented in the same
process) with respect to the lowest supply voltage required for sustained oscillations (𝑉DDmin ), oscillation
frequency and peak-to-peak voltage swing (𝑉pp ). For a supply voltage of 50 mV, over 90% 𝑉DD is obtained
from post layout simulations for all four proposed architectures consuming around 24 pW of average power,
out of which the variant that has the maximum swing of 92% (an improvement of 9.2% compared to that of
stacked inverter based RO), is able to start and sustain an oscillation at 32.5 mV supply voltage. The fastest
architecture proposed has a frequency (post layout) of 131.5 Hz at 50 mV 𝑉DD , which is 62% more than that
of the stacked inverter based RO with body bias. Monte Carlo analysis reveals that the proposed RO variants’
𝑉pp have lesser interquartile range (IQR) and relatively higher median values.
∗ Corresponding author.
E-mail addresses: ankur.mkjee@greenpmusemi.com (A. Mukherjee), saichandrateja@greenpmusemi.com (Sai Chandra Teja R.).
https://doi.org/10.1016/j.mejo.2023.105883
Received 5 April 2023; Received in revised form 26 June 2023; Accepted 26 June 2023
Available online 3 July 2023
0026-2692/© 2023 Elsevier Ltd. All rights reserved.
A. Mukherjee et al. Microelectronics Journal 139 (2023) 105883
Table 1
Body connection for each RO configuration.
Body bias scheme BpST𝑖 BnST𝑖 BpM𝑖 BnM𝑖 BpSB𝑖 BnSB𝑖
NN-NN-NN IN𝑖+2 IN𝑖+2 IN𝑖+2 IN𝑖+2 IN𝑖+2 IN𝑖+2
XX-NN-YY X𝑖+1 X𝑖+1 IN𝑖+2 IN𝑖+2 Y𝑖+1 Y𝑖+1
SBBIRO VDD VSS IN𝑖+2 IN𝑖+2 VDD VSS
XX-XY-YY X𝑖+1 X𝑖+1 X𝑖+1 Y𝑖+1 Y𝑖+1 Y𝑖+1
SIRO VDD VSS VDD VSS VDD VSS
XX-GS-YY X𝑖+1 X𝑖+1 VSS VDD Y𝑖+1 Y𝑖+1
2
A. Mukherjee et al. Microelectronics Journal 139 (2023) 105883
Fig. 3. Two adjacent delay cells (𝑖th and 𝑖 + 1th stages) in the RO from Fig. 2 showing
the nomenclature of the devices and the body terminals.
Table 2
Delay cell device dimensions.
Device Width Length
MpST 13 μm 0.18 μm
MnST 1.3 μm 0.4 μm
MpM 5 μm 0.18 μm
MnM 0.9 μm 0.4 μm
MpSB 8 μm 0.18 μm
MnSB 4 μm 0.4 μm
Table 3
Layout area comparison.
RO type Dimensions in μm Area in μm2
NN-NN-NN 125 × 47 5875
XX-NN-YY 192 × 60 11 520
SBBIRO 210 × 60 12 600
XX-XY-YY 210 × 60 12 600
SIRO 192 × 60 11 520
XX-GS-YY 210 × 60 12 600
The cells are designed and optimized in the 180-nm tech node using
regular 𝑉𝑡 devices (of approximately 500 mV), although low or medium
𝑉𝑡 devices could be used to obtain higher sub-threshold currents. The
device dimensions are tabulated in Table 2. The middle inverter was
sized first to obtain a decent gain and symmetric voltage transfer
characteristic (VTC).
MpST (MnSB) are sized much wider than MpM (MnM) so that the
overall pull-up (pull-down) strength of MpM (MnM) is not affected by
the stacked top (bottom) inverters (also mentioned in [7]).
Having lesser pull-down (pull-up) strengths for MnST (MpSB) com-
pared to the pull-up (pull-down) strengths MpST (MnSB) is necessary so
that these devices do not interfere with the pull-up (pull-down) action
of MpST (MnSB), however, weakening them too much would elevate
the leakage current. With increase in leakage current, the 𝑖on /𝑖off ratio
drops thereby bringing down the overall swing at ultra-low voltages.
Fig. 4. Schematics of two adjacent delay cells (𝑖th and 𝑖 + 1th stages) in the RO from
Also, to fix the number of stages 𝑛, the time period of oscillation needs
Fig. 2 of all the proposed variants.
to be sufficiently greater than the rise and fall time of the delay cell.
This is to ensure that the nodes in the RO chain are able to reach and
settle at their minimum/maximum value so as to obtain a rail-to-rail
square pulse with maximum 𝑉pp . The layouts of four 13-staged ring Fig. 5. The layout area is the least for the NN-NN-NN variant, roughly
oscillators using the proposed four variants of delay cells are shown in half the area compared to the other variants (Table 3), as all the body
3
A. Mukherjee et al. Microelectronics Journal 139 (2023) 105883
Fig. 6. Output voltage (IN𝑖+1 ) and gain versus input voltage (IN𝑖 ) applied to all the
delay cell variants.
4
A. Mukherjee et al. Microelectronics Journal 139 (2023) 105883
5
A. Mukherjee et al. Microelectronics Journal 139 (2023) 105883
Table 5
Peak-to-peak voltage swing and minimum supply voltage required for sustained
oscillations for all types.
RO type 𝑉pp at 𝑉DD = 50 mV 𝑉DDmin
NN-NN-NN 46 mV 32.5 mV
XX-NN-YY 45.45 mV 34 mV
SBBIRO 42.56 mV 37 mV
XX-XY-YY 45.23 mV 35 mV
SIRO 41.41 mV 38 mV
XX-GS-YY 45.02 mV 36 mV
Fig. 11. Frequency versus supply voltage for different RO Architectures showing
variation across the fast and slow corners.
6
A. Mukherjee et al. Microelectronics Journal 139 (2023) 105883
Fig. 12. Number of delay cells in the RO chain versus frequency for different RO
Architectures at a supply voltage of 100 mV.
Fig. 14. Average current taken by the body terminals in a single delay cell for different
supply voltages.
Four new variants of a body biased stacked inverter based delay cell The authors declare the following financial interests/personal rela-
are proposed using which 13-staged ring oscillators (for each variant) tionships which may be considered as potential competing interests:
are implemented in 180-nm BCD CMOS process, for which post layout Prof Ashudeb Dutta reports financial support was provided by WxBunka
simulations are carried out. The variants are compared with the pre- Foundation.
vious works, SI and SBBI based RO, both of which being implemented
using the same unit delay cell (by appropriately connecting the body Data availability
terminals as mentioned in the works as discussed above) in the same
process along with the other four. All the architectures implemented No data was used for the research described in the article.
consume around 24 pW of average power at 𝑉DD of 50 mV. Over 90%
𝑉pp at 50 mV 𝑉DD is achieved in all the proposed RO types the highest Acknowledgement
being 92% for NN-NN-NN which is 6.9% and 9.2% more than that of
SBBIRO and SIRO respectively. The same variant (NN-NN-NN) has the This research was supported by WNI WxBunka Foundation, Japan.
maximum gain per unit cell and is able to start and sustain oscillation The authors acknowledge Prakash Goriparthi and Raviteja Sanampudi
for a supply voltage as low as 32.5 mV; which is lower than the SBBIRO for the layout design.
7
A. Mukherjee et al. Microelectronics Journal 139 (2023) 105883