130nm SiGeHBT技术实现的超低功耗200GHz放大器

需积分: 9 167 浏览量更新于2024-07-15 收藏 114.17MB PDF 举报

"这篇论文详细介绍了使用130纳米硅锗双极晶体管（SiGe HBT）技术设计的一种超低功耗173至207 GHz放大器。该放大器由三个级联的增益单元组成，在实验性的130纳米SiGe HBT工艺中实现，该工艺的峰值fT/fmax达到460/600 GHz。为了实现极其低的静态直流功率消耗，电路设计使得晶体管在正向偏置基发射结电压（VBC）下工作。在1.3伏的电源电压下（VBC约为0.2伏），放大器在180和205 GHz时表现出23.5分贝的峰值增益，具有从173到207 GHz的34 GHz 3-dB带宽。即便在0.7伏的电源电压下（VBC约为0.5伏），放大器仍能在175 GHz时保持18.3分贝的峰值增益，功耗仅为3.2毫瓦。" 这篇文章是《IEEE固态电路杂志》（JSSC）2020年6月刊的一部分，主要探讨了固态电路领域的最新研究成果，特别是针对集成电路的晶体管级设计。在这个研究中，研究人员提出了一种创新的、低功耗的毫米波放大器设计，它在超高的频率范围内工作，如173至207 GHz。使用的器件基于130纳米的硅锗异质结双极晶体管技术，这是一种高速半导体技术，适合高频电路应用。放大器的架构由三个串联的增益单元组成，这种级联设计有助于提高增益和带宽。关键在于设计电路时考虑了晶体管在正向偏置基发射结电压下操作，这有助于降低静态直流功率损耗。在1.3伏的供电电压下，放大器的性能非常出色，增益达到了23.5分贝，同时保持了相当大的带宽，从173 GHz扩展到207 GHz，3-dB带宽为34 GHz。即使在较低的0.7伏电压下，放大器仍然能够在175 GHz处提供18.3分贝的增益，表明了其在不同工作条件下的鲁棒性。这个研究不仅展示了高性能的毫米波放大器设计，还强调了通过优化晶体管工作状态来实现低功耗的方法，这对于便携式和远程通信系统，尤其是需要在极高频率下工作的无线设备，具有重要意义。这种技术进步可能对未来的射频和微波电路设计产生深远影响，特别是在5G通信、卫星通信以及雷达系统等领域。

1702 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 55, NO. 6, JUNE 2020

Fig. 5. (a) IL-RO-based F2A converter and the output wave at each node. (b) Illustrations of output wave at nodes in the F2A converter when the IL-RO is

locked/pulled by a signal with f

inj

> f

or f

inj

< f

. (c) Output amplitude of the F2A converter against the injection signal frequency.

tones with most of the power concentrated at f

inj

and f

∗

[30]

∗

= f

inj

∓



( f

inj

− f

)

− f

(2)

inj



⎛

⎝

inj

− f

−





inj

− f



− 1

⎞

⎠



· P (3)

= 4





inj

− f



− 1



⎛

⎝

inj

− f

−





inj

− f



− 1

⎞

⎠

−2

· P. (4)

where f

is the IL range determined by the injection strength

inj

osc

) in one IL-RO. P

inj

, P

,andP are the power spec-

tral densities at f

inj

, f

∗

and total output power, respectively.

Similarly, the AM-modulated RO output during injection

pulling after the ED and LPF is shown in Fig.5 (a) and (b).

To simplify the analysis and relate the amplitude with

frequency in a similar manner as (1), we deﬁne an equivalent

frequency f

as follows:

⎡

⎣

⎛

⎝



i=1,2,3...

∗

⎞

⎠

inj

⎤

⎦

(5)

where f

∗

is the frequency of the i th tone of the IL-RO output

under injection pulling state, P

is its corresponding power

spectral density, and P = (



i=1,2,3...

)+P

inj

. As mentioned

earlier, P

inj

and P

would contribute the largest portion of

converter output power.

At the injection pulling state, f

in (5) is actually a

weighted sum of power spectral densities of all tones with its

value lies within ( f

− f

, f

+ f

). It can indicate the state

of injection pulling. As injection signal frequency deviates a

lot from f

,i.e., f

goes larger, the injection pulling effect

gets weaker, and f

approaches f

. Using the equivalent

frequency concept, the ﬁltered envelope output during pulling

state can be formulated similarly as (1)

A = K ·



−



+ A

= K ·

⎡

⎣

⎛

⎝



i=1,2,3...

∗

inj

⎞

⎠

−

⎤

⎦

+ A

. (6)

As a result, compared to the converter’s output without

injection, the converter would output a smaller/larger ampli-

tude during injection pulling when f

is higher/lower than f

As shown in Fig. 5(c), using an injection strength of 0.1,

the calculations based on (1)–(6) match well with the sim-

ulation result. We observe that the converter equivalent

frequency f

approaches the free-running frequency f

and

the converter output approaches its intrinsic amplitude at free

running when the injection frequency deviates a lot from f

which indicates the weakening of the pulling effect.

Authorized licensed use limited to: ShanghaiTech University. Downloaded on May 30,2020 at 12:20:29 UTC from IEEE Xplore. Restrictions apply.

XIONG et al.: 7 ×7 ×2 mm

8.6-µW 500-kb/s TX WITH ROBUST IL 1703

Fig. 6. (a) Proposed F2A converters-based demodulator. (b) Proposed robust

demodulation based on two F2A converters.

B. Frequency Variation Resilient Demodulation

By exploiting the F2A characteristic, a frequency variation

resilient demodulator [31] is proposed, as shown in Fig. 6(a).

The proposed demodulator consists of two IL-RO-based

F2A converters, with its two ROs free-running frequency

centered at f

and f

, respectively. Under the injec-

tion strength of 0.1, the two measured ﬁltered outputs

and O

) against the injection frequency will exhibit ﬁrst-

order derivative of Gaussian pulse characteristic (blue and red

curves), as shown in Fig. 6(b), centered at their respective

self-resonant frequencies ( f

and f

). In addition, the two

curves deviate slightly from each other along the ampli-

tude axis. The red curve with higher self-resonant frequency

( f

) is shifted downward compared with the blue curve

with lower self-resonant frequency ( f

). This is understand-

able given that lower self-resonant frequency allows longer

charging/discharging period, and thus results in higher ﬁltered

amplitude. Furthermore, both curves exhibit voltage variation

of more than 100 mV under different injection frequencies.

By employing a comparator (CMP) with resolution smaller

than V

, any injected TX frequency that falls within the light

blue region will result in a negative difference between the two

F2A converters output (O

−O

<0). Similarly, any injected

TX frequency that lies within the light red region will result in

a positive difference (O

−O

>0). As illustrated in Fig. 6(b),

the relatively wide light blue and light red regions imply that

the demodulator would be able to handle implanted TX with

large carrier ﬂuctuation. Unlike other reported IL-F2As, the

proposed IL-F2A-based demodulator targets at implanted TX

with inaccurate frequency generation and potential frequency

ﬂuctuations arise from chip-to-chip variation. This allows the

implanted TX to eliminate accurate crystal oscillator reference

and achieve miniaturization and low power.

V. C

IRCUIT IMPLEMENTATION

A. Implanted TX Radio Block

The implanted TX consists of PMU and TX radio block, as

shown in Fig. 7. The on-chip PMU converts the micro-battery

output into a regulated 0.8 V needed by the TX radio

block. The on-chip PMU eliminates the potential frequency

Fig. 7. Schematic of the implanted TX.

Fig. 8. Schematic of the on-chip PMU in the implanted TX.

ﬂuctuations due to supply and minimize the external com-

ponent counts. Two three-stage free-running PMOS-loading

differential ROs, two differential-to-single-ended converters,

and two tri-state buffers constitute the desired reference-less

FSK generator. The differential design is adopted to achieve

better common mode and supply noise rejection. Manchester

coding is adopted to embed the data clock information into

baseband data stream for clock recovery at RX side [32].

A 200-fF impedance matching network is employed to drive

the large micro-probe impedance, which further improves the

output buffer efﬁciency.

B. PMU in Implanted TX

On-chip PMU serves two main purposes. First, it regulates

the supply to the TX radio block and minimizes the potential

frequency ﬂuctuation due to supply. Second, it eliminates

external component count and helps in miniaturizing the

implanted TX. The commercially available micro-batteries

exhibit an output voltage range from 2.8 to 4.2 V [33]. A Seiko

MS412FE micro-battery with an output voltage of 3.2 V is

used in this article. Its capacity is 1 mAh with a maximum

load current of 100 µA [34]. For low-power operation, it is

desirable to power the TX radio block under 0.8 V. The PMU

consists of a typical dc–dc converter and an low drop-out

regulator (LDO), as shown in Fig. 8.

Authorized licensed use limited to: ShanghaiTech University. Downloaded on May 30,2020 at 12:20:29 UTC from IEEE Xplore. Restrictions apply.

1704 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 55, NO. 6, JUNE 2020

To improve the PMU efﬁciency, a 1/3-switched-cap dc–dc

converter is ﬁrst used to convert the 3.2-V battery voltage into

about 1.1 V before passing it to LDO. The LDO then regulates

the dc–dc output to 0.8 V, outputting a current of 10.5 µA

while maintaining the output supply ripple to within 1%. The

dc–dc converter is switched by a clock signal derived from

one RO within the TX FSK generator. The clock frequency is

obtained by passing the RO through a 1/64 divider. 500 kHz is

chosen for the resulting clock signal (CLK_SW) as well as the

baseband clock (CLK_DATA) to achieve the desired 500-kb/s

data rate. By reusing RO within the FSK generator for clock

generation, it reduces the overall PMU power consumption

after startup.

Although the PMU architecture is straightforward, there

exists a startup issue that deserves some discussion. This issue

arises due to reusing RO for clock generation to achieve power

saving. This is because RO is powered by PMU, which might

not be functioning if the switching clock is not available.

Hence, an auto clock switching scheme is adopted here to

eliminate this startup issue. During the startup phase, a simple

inverter-based RO directly powered by the micro-battery will

be used to drive the PMU with an oscillation frequency

of 200–300 kHz. Lower frequency is chosen here to reduce

the startup power consumed by the PMU. This clock will

be used by the dc–dc converter until the LDO output is

sufﬁcient to power up the ROs within the FSK generator.

In our PMU design, once the LDO output attains 0.5 V,

a level detector will automatically switch the clock signal

from micro-battery-powered RO to the RO within the FSK

generator. To reduce area and power of PMU, two PMOS

pseudoresistors in series are employed to generate a reference

voltage, which is half of TX Vdd. This reference voltage

will drive a resistive load inverter (P1 and P2). By sizing

P1 and P2 appropriately, the inverter will switch when TX

Vdd reaches 0.5 V. The level detector output, CTRL, will then

toggle the PMU clock signal from micro-battery-powered RO

to CLK_SW automatically. Once the auto clock switching is

done, the startup RO will be powered down to save power.

In the PMU, a ULP VTH reference [35] will provide a

supply insensitive 400-mV reference to the LDO. It will also

be used to generate bias needed for all blocks within the TX.

A charge pump is employed to drive the switched-cap network.

It converts the periodic 500-kHz CLK_SW signal at low volt-

age (0–0.8 V) into full micro-battery voltage level (0 ∼ 3.2 V)

while consuming less than 85 nW. Non-overlapping clock

generator is also used to avoid simultaneous turning on of

the switches.

C. On-Body RX

As mentioned earlier, the lack of accurate reference clock

for the implanted TX results in signiﬁcant frequency ﬂuctua-

tion. To overcome this issue, IL-F2A-based frequency varia-

tion resilient demodulator discussed earlier is shown in Fig. 9.

In the RX, the received signal from the two on-body

electrodes is ﬁrst ampliﬁed by a three-stage bandpass front-end

ampliﬁer (FEA) with 50-dB gain. An on-chip balun cou-

pled with inverters converts the single-ended signal into

pseudo-differential full swing signal to minimize the common

Fig. 9. Schematic of the on-body RX.

mode noise and enlarge the dynamic range. As the on-body

electrodes within the galvanic-couple I2O-BCC band exhibits

relatively high impedance, a simple self-biased inverter archi-

tecture is adopted as the FEA. The same current-starved

pseudo-differential inverter architecture is adopted in both

the injection and delay cells. The injection strength can be

tuned by digitally adjusting the tail transistor sizing ratio. The

desired injection strength of 0.1 is determined based on the

simulation and measurement. This value will generate two

wide regions (light blue and light red in Fig. 6) for proper

demodulation of the two TX FSK frequencies despite their

potential large variations. The two three-stage current-starved

pseudo-differential ROs are free running at different frequen-

cies. Their center frequencies should be adjusted roughly to

fall within the I2O-BCC band. A simple differential rectiﬁer

coupled with RC ﬁlter is employed as ED. Additional second-

order LPF is used to extract the average value of the two

IL-F2A outputs. The ED exhibits conversion gain around 0 dB,

a bandwidth of 1.5 MHz, and output noise less than 400 µV

To achieve a minimum SNR of 20 dB, the ED output voltage

needs to be at least 4 mV

. By taking into TX output voltage,

the channel loss, and the ED output voltage requirement,

the desired FEA gain can be found. The LPF-dB gain is

with bandwidth of 1.5 MHz. The LPF output corresponding

to the two IL-F2As is then sent to a differential regenerative

current feedback Schmitt trigger as shown in Fig. 6(b). The

Schmitt trigger is designed to have a hysteresis of 300 mV. The

amplitude difference between the two IL-F2A outputs can be

adjusted by tuning the RC value of the ED, the common mode

of the LPF, and the input offset of the Schmitt trigger. The

demodulated data will then pass through a Manchester decoder

to extract both the baseband data and clock information.

VI. M

EASUREMENT RESULTS

Fabricated in Global Foundries 130-nm CMOS technology,

the in-body TX and on-body RX die photos are shown

in Fig.10.

A. Implanted TX

As the TX with on-chip PMU will be implanted in human

body, we expect the major contribution to the FSK frequency

variation would be from the process variation. It should

be noted that to reduce the cost, batch trimming should

Authorized licensed use limited to: ShanghaiTech University. Downloaded on May 30,2020 at 12:20:29 UTC from IEEE Xplore. Restrictions apply.

XIONG et al.: 7 ×7 ×2 mm

8.6-µW 500-kb/s TX WITH ROBUST IL 1705

Fig. 10. Die photographs of implanted TX and on-body RX.

Fig. 11. Measured two ROs’ frequencies in TX and f

distribution among

20 chips before and after 1-point batch trimming.

be applied, i.e., identical setting to be applied to all TXs.

In addition, the main objective of the trimming is to center the

FSK signals within the desired frequency band (10–40 MHz)

where the galvanic coupling exhibits optimal channel charac-

teristics. Before trimming, among the 20 measured TX chips

(Fig. 11), the two TX ROs’ frequencies ( f

FSK_2

and f

FSK_1

)

are separately centered at 41.3 and 36.1 MHz with standard

deviation (δ

FSK_2

and δ

FSK_1

) less than 1.31 MHz. At the

same time, the observed average frequency difference ( f

)

between the two FSK tones is 5.17 MHz with standard

deviation (δ

) of 0.89 MHz. After applying 1-point batch

frequency trimming, the resulting two TX ROs’ frequen-

cies ( f

FSK_2

and f

FSK_1

) are separately centered at 32.5 and

24.6 MHz with δ

FSK_2

and δ

FSK_1

less than 1.03 MHz, and the

observed average f

of the FSK is 7.71 MHz with 0.92 MHz

variation. From the measurement, the typical case, upper limit

case, and lower limit case of f

for the TX FSK signals

are7.7MHz(f

FSK_2

= 32.2MHzand f

FSK_1

= 24.5MHz),

9.9 MHz ( f

FSK_2

= 34.5MHzand f

FSK_1

= 24.6MHz),and

6.0 MHz ( f

FSK_2

= 31.2MHzand f

FSK_1

= 25.2MHz),

respectively.

In the overall system measurement, the implanted TX

can generate 500-kb/s FSK signals with 30-mVpp output

amplitude at the two micro-probes within pork (Fig. 13 right

Fig. 12. Startup of the PMU in implanted TX and its efﬁciency.

above), achieving energy efﬁciency of 17.2 pJ/b. The speciﬁc

safety absorption rate (SAR) of the proposed system is less

than 0.4 µWrms, which satisﬁes the safety requirements

of the International Commission on Non-Ionizing Radiation

Protection (ICNIRP) guidelines with sufﬁcient margins for

apersonweighting50kg(<100 Wrms) [36]. In addition,

the current density from the proposed TX probes also meets

the ICNIRP guidelines, which requires current density through

the human body to be below 25 mA/m

over a frequency range

of 100 kHz–60 MHz [37].

For the on-chip PMU within the implanted TX, its measured

output regulated voltage is 800 mV with error smaller than

1.4% when the battery voltage is above 2.8 V. At 3.2-V

battery voltage, it exhibits a power efﬁciency of 45%. The

startup phase and clock signal auto-switching are also shown

in Fig. 12. As illustrated, the PMU will go through four states

to attain the stable regulated output voltage. During state 1,

startup RO directly powered by battery is employed to drive

the PMU. Once the LDO output reaches 0.5 V, clock signal

auto-switching is activated. However, the RO within TX FSK

generator still has limited output swing and might not be able

to sustain the PMU operation. Hence, we observe the back

and forth clock switching between the two ROs, as indicated

by the CTRL pulses (yellow) during state 2. LDO output still

keeps increasing despite the constant clock signal switching.

Once the LDO output reaches 600 mV, the PMU will enter

state 3 with stable CTRL of “1.” The dc–dc converter output

and LDO rise steadily to about 1.1 V and 800 mV, respectively.

The PMU now enters state 4, which can now provide stable

regulated output voltage for the implanted TX.

Due to the low quiescent current of the PMU and the

low power TX radio block, the current drained from the

battery during normal operation is less than 10 µA, which

is 10× lower than the peak current allowable by the micro-

battery. Hence, the overall implanted TX can last for 1 week

Authorized licensed use limited to: ShanghaiTech University. Downloaded on May 30,2020 at 12:20:29 UTC from IEEE Xplore. Restrictions apply.

1706 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 55, NO. 6, JUNE 2020

TABLE I

ERFORMANCE COMPARISON WITH PRIOR WORKS

and be directly powered by the micro-battery without any other

external components [5].

B. On-Body RX

Based on the trimmed f

FSK_1

and f

FSK_2

distributions

of TX, f

and f

in RX are then coarsely tuned to center

at 28.2 and 37.0 MHz. The main objective of the trimming is

again to ensure the transmission occurs at the optimal channel

band (10–40 MHz). With an off-chip regulator, the expected

and f

variation is less than 5% over a temperature

range of 0

◦

C–80

◦

C as indicated by the narrowly shaded

regions surrounding f

and f

in Fig. 6(b), which only

constitutes a small portion of the light blue region. Given

1-point TX batch trimming, large frequency variation is still

observed for the 20 chips. However, with the proposed IL-F2A

demodulator, proper demodulation with measured bit error rate

(BER) < 0.1% can be achieved as long as the TX observes

FSK_1

< 27.4MHzand29.0MHz< f

FSK_2

< 36.0MHz.

In summary, TX FSK frequency variations of more than 20%

can be tolerated with the proposed architecture.

C. Overall System

To verify the feasibility of the proposed I2O-BCC com-

munication, we packaged our in-body TX, micro-probes, and

micro-battery together on the DuPont Kapton Polymide ﬁlm

with 25 µm thickness, and coated with parylene-C except for

the two micro Pt/Ir probes. The complete packaged implanted

TX measured only 7 ×7 ×2mm

and is mainly limited by

the micro-battery size and the adopted system-in-package tech-

nology. The full system setup and fully packaged implanted

TX are shown in Fig. 13. To demonstrate the wireless setup

in WBAN, the recovered baseband data and clock through

on-body RX are sampled by Raspberry Pi Zero W and

transmitted off-body via its Wi-Fi module. The TX radio block

Fig. 13. (Left above) Fully packaged implanted TX. (Left below) The

measurement of the I2O-BCC system. (Right above) The measured implanted

TX output wave. (Right below) The measured BER of on-body RX at three

TX FSK cases.

consumes only 8.6 µW to send the FSK signal into tissue. The

total power consumption of the implanted TX, including the

PMU, is 19.2 µW. Powered by micro-battery (3.2 V, 1 mAh),

the whole implanted TX can operate continuously for 1 week.

Fig. 14 shows the successful demodulation for FSK signals

for three f

cases from TX. Both the clock and baseband

data recovery for three f

cases are demonstrated. The pro-

posed RX can achieve data rate of 500 kb/s with a sensi-

tivity of 300 µV at 0.1% BER for all three cases in the

overall system measurement. This is consistently achieved

with in-body-to-on-body distance of 10 cm over a measure-

ment period of 48 h (limited mainly by measurement setup).

This clearly indicates the feasibility and robustness of the

proposed transceiver.

The performance and comparison with recent implanted

TXs and their corresponding external RXs [5], [8], [9], [11],

[38]–[41] are summarized in Table I. In our system, the TX

power consumption is at least 8× lower than the prior

Authorized licensed use limited to: ShanghaiTech University. Downloaded on May 30,2020 at 12:20:29 UTC from IEEE Xplore. Restrictions apply.

剩余302页未读，继续阅读

netshell

粉丝: 11
资源: 185

130nm SiGeHBT技术实现的超低功耗200GHz放大器

JSSC_202001.pdf

JSSC_202012.pdf

JSSC_202004.pdf

JSSC_202009.pdf

JSSC_202005.pdf

JSSC_202007.pdf

15_sar.pdf

IEEE JSSC 202009: Solid-State Circuits and Integrated Circuits Special Section

S变换+Sockwell R G , Mansinha L , Lowe R P . Localization of the complex spectrum: the S transformJ

2021科大讯飞车辆贷违预测大赛冠军源码+全部资料.zip

最新资源